how capital gains tax rates affect revenues: the historical

CONGRESS OF THE UNITED STATESCONGRESSIONAL BUDGET OFFICE

How Capital Gains TaxRates Affect Revenues:The Historical Evidence

MARCH 1988

ACBO STUDY

HOW CAPITAL GAINS TAX RATES AFFECTREVENUES: THE HISTORICAL EVIDENCE

The Congress of the United StatesCongressional Budget Office

PREFACE

The taxation of capital gains has been changed frequently in the past two decades,most recently in the Tax Reform Act of 1986. What have been the effects of thesechanges on revenues? The revenue effects of changes in capital gains taxation areuncertain because taxpayers may choose to hold onto their assets instead of sellingthem. Thus, changes in realizations of gains can offset the direct effects of changesin tax rates. This report reviews previous research on the effects of changes incapital gains tax rates on the realizations of gains and presents new evidence basedon a statistical analysis of taxpayer behavior over the past 30 years. The new esti-mates of behavioral responses are used to assess the probable revenue consequencesof the recent Tax Reform Act and of a proposal to lower the maximum tax rate oncapital gains. The report was prepared in response to separate requests from Con-gressman William Gray III, Chairman of the House Budget Committee, and Con-gressman Willis Gradison, member of the House Budget Committee and HouseWays and Means Committee. In keeping with the mandate of the CongressionalBudget Office to provide objective analysis, the study contains no recommendations.

The paper was prepared by Eric Toder and Larry Ozanne of the Tax AnalysisDivision, under the direction of Rosemary Marcuss. A number of people inside andoutside of CBO reviewed drafts and provided valuable comments. They includeGerald Auten, Thomas Barthold, Leonard Burman, Paul Courant, Albert Davis,Frank deLeeuw, Marilyn Flowers, Edward Gramlich, Jane Gravelle, Jon Hakken,Eric Hanushek, Richard Kasten, Donald Kiefer, John O'Hare, Rosemarie Nielsen,Rudolph Penner, Jack Rodgers, Frank Sammartino, Joel Slemrod, Ralph Smith,John Sturrock, and Bruce Vavrichek. Responsibility for the finished product,however, rests with CBO. The revenue simulations using the CBO IndividualIncome Tax Model were performed by Frank Sammartino. Daniel Polsky providedcomputational assistance. Francis Pierce edited the manuscript. Linda Brockmanprepared early drafts of the manuscript, and Kathryn Quattrone prepared the finaldraft for publication.

James L. BlumActing Director

March 1988

CONTENTS

SUMMARY xi

I INTRODUCTION 1

H PREVIOUS RESEARCH ON THEREVENUE EFFECT OF CAPITALGAINS TAXES 5

Approaches to Estimating theRealizations Response 5

Measures of Revenue Effects 7The Studies 8Other Revenue Effects of

Capital Gains Tax Changes 17

m CAPITAL GAINS TAXES ANDREALIZATIONS, 1954-1985 21

Distribution of Gains AmongIncome Groups 21

Growth in Capital GainsRealizations Over Time 24

Marginal Tax Rates and Realizationsof Gains: A First Glance 40

Revenue from Capital Gains Taxes 41

IV STATISTICAL EVIDENCE ONCAPITAL GAINS ANDTAX REVENUES 45

Determinants of CapitalGains Realizations 45

Statistical Estimates 51Reliability of the Estimates

and Their Implications forRevenue Effects 59

Evaluation of Findings 66

vi HOW CAPITAL GAINS TAX RATES AFFECT REVENUES March 1988

APPENDIXES

A Alternative Estimates of theResponse of Capital GainsRealizations to Tax Rates 73

B Methodological Issues in Studiesof the Realizations Response 9 7

CONTENTS vii

TABLES

1. Summary of Previous Studies on RevenueEffects of Capital Gains Taxes 18

2. Distribution of Long-Term Capital Gainsby Adjusted Gross Income (AGI): 1984 22

3. Realized Net Long-Term Gains Comparedwith Gross National Product, 1954-1985 25

4. Realizations of Net Long-Term GainsCompared with Corporate Equity Heldby Individuals, 1954-1985 27

5. Ratio of Realized Net Long-Term Gainsto Gross National Product, by IncomeGroup, 1954-1985 29

6. Share of Realizations of Net Long-TermGains by the Highest-Income Groups,1954-1985 31

7. Maximum Marginal Tax Rate onCapital Gains, 1954-1988 36

8. Estimated Weighted Average MarginalTax Rate on Capital Gains, 1954-1985 38

9. Revenue Attributable to Taxes onLong-Term Capital Gains, 1954-1985 43

10. Regression Equations Explaining CapitalGains Realizations for All Taxpayers,1954-1985 52

11. Actual Minus Fitted Gains in Recent Yearsin Equations for All Taxpayers 54

12. Regression Equations Explaining CapitalGains Realizations for the Top 1 Percentof Returns, 1954-1985 56

viii HOW CAPITAL GAINS TAX RATES AFFECT REVENUES March 1988

13. Regression Equations Explaining CapitalGains Realizations for the Bottom 99 Percentof Returns, 1954-1985 57

14. Confidence Intervals for Tax Effect onRealizations and Revenue-MaximizingRate for Entire Population of Taxpayers 60

15. Implied Shape of Relationship Between TaxRates and Revenue from Capital Gains Taxes 61

16. Comparison of Capital Gains Revenue UnderPre-1986 Law with Revenue Under CurrentLaw and Under a 15 Percent Maximum Rate 64

A-l. Alternative Equations ExplainingCapital Gains Realizations 76

A-2. Equations Explaining Capital GainsRealizations Using Wealth Instead ofIncome Variables 83

A-3. Equations Explaining Capital GainsRealizations Using Alternative TaxRate Specifications 88

A-4. Equation Explaining Capital GainsRealizations Using Changes in AssetValues and Basis 91

CONTENTS

FIGURES

1. Realized Net Long-Term Gains and GrossNational Product, 1954-1985 26

2. Realized Net Long-Term Gains andCorporate Equity of Households, 1954-1985 28

3. Ratio of Realized Long-Term Gains to GrossNational Product, by Income Group, 1954-1985 30

4. Average Marginal Tax Rates on Long-TermGains for Selected AGI Groups 39

5. Marginal Tax Rates on Long-Term Gainsand the Ratio of Long-Term Gains toGross National Product 41

6. Relationship of Revenue from CapitalGains Taxes to Marginal Tax Rates on Gains 62

A-l. Relationship of Revenue from CapitalGains Taxes to the Marginal Tax Rate onGains for Four Specifications of theTax Rate Variable 90

BOXES

1. Estimating Accrued Gains 47

2. Revenue Simulations 65

SUMMARY

The Tax Reform Act of 1986 lowered marginal personal income taxrates but also eliminated many tax preferences, including the 60 per-cent deduction for long-term capital gains. The maximum tax rate onlong-term gains was increased from 20 percent under previous law to28 percent for the highest-income taxpayers and 33 percent fortaxpayers just below the highest-income group. The reasons for elimi-nating the capital gains deduction were to help finance the reductionin ordinary income tax rates, to allow the top rate to be cut substan-tially without providing disproportionate relief to the highest-incomegroup, and to simplify the tax system.

How much additional revenue will be obtained by increasing taxrates on capital gains is uncertain. Taxpayers can defer the paymentof capital gains taxes by not realizing the gains-that is, by holding on-to assets instead of selling them; they can avoid taxation of gains en-tirely by passing on their assets to others at death. If realizationsdecline by a greater percentage than the tax rate increases, revenuesfrom capital gains taxes could fall instead of increasing.

A number of statistical studies have provided strong evidence thatrealizations of capital gains decline when tax rates on gains areincreased. The estimated size of this response of capital gains realiza-tions, however, differs greatly among studies. The responses esti-mated in some studies have been used to support a claim that the 1986act reduced revenue from capital gains taxes when it increased the taxrates, and that lowering the maximum tax rate on long-term gains to15 percent would increase revenue. The estimates in other researchsuggest an opposite conclusion. Moreover, all of these studies haveused methodologies that are open to criticism.

This study provides new evidence on the relationship betweenrealizations of long-term capital gains and tax rates on capital gains,based on statistical analysis of data for the years 1954 through 1985.The statistical results offer additional support for the view that highertax rates do lower realizations of capital gains. As a result, increasesin tax rates on capital gains produce much less revenue than theywould if taxpayers' behavior were unaffected. On the other hand,

xii HOW CAPITAL GAINS TAX RATES AFFECT REVENUES March 1988

simulations using the estimated behavioral responses still show a netrevenue increase from the 1986 act. They also indicate that loweringthe top rate on long-term capital gains to 15 percent would result in anet revenue loss.

The estimates of the behavioral response contain considerable sta-tistical uncertainty. The proposition that a maximum tax rate of 15percent would yield more revenue than current law rates cannot beruled out with certainty, although the probability attached to thisresult is very low. Similarly, the proposition that revenue from capi-tal gains taxes is maximized at rates far above those of current lawalso cannot be ruled out.

This report is concerned only with the issue of estimating revenue.Many other factors need to be considered in deciding how to tax capitalgains. Arguments for lower tax rates on gains are that they promotesaving and investment and channel more resources into new ventures.In addition, a preferential rate on nominal gains provides a rough ad-justment for the fact that some gains reflect inflation instead of realincreases in purchasing power (though one could directly eliminatethe taxation of the inflationary component of gains without intro-ducing a preferential rate). Arguments against reintroducing a differ-ential between long-term and short-term capital gains and ordinaryincome by lowering the tax rate on capital gains are that the dif-ferential would add complexity to the tax system, encourage tax shel-ter activity, and distort choices among financial instruments and realassets.

PREVIOUS RESEARCH ON REVENUEFROM CAPITAL GAINS TAXES

Previous studies have estimated how much taxpayers change theircapital gains realizations in response to a change in the tax rate onthose realizations. Those studies use two different approaches: thecross section and the time series. Cross-section studies compare thebehavior of taxpayers or taxpayer groups in the same year or overseveral years. They examine the effect that differences in marginaltax rates among taxpayers have on differences in their capital gainsrealizations, controlling for the effects of other influences such as

SUMMARY xiii

dividends, total income, age, and family status. The estimate of theeffect of differences in marginal tax rates on differences in realizationsis then used to infer how taxpayers would respond if tax rates oncapital gains were changed. In contrast, time-series studies examinethe effect of differences in marginal tax rates over time on total reali-zations of gains, again controlling for other influences on realizedgains such as real income, wealth, and the price level.

The studies also differ among each other in the data samples theyuse, the way they define marginal tax rates, the set of other variablesincluded as determinants of gains realizations, and the way theyadjust for particular statistical problems. Consequently, they havefound a wide range of responses, with very diverse implications for therevenue effects of changing the tax rate on capital gains. The onegenerally common finding in all of the studies is that higher marginaltax rates lower realizations. In several of the cross-section studies,and in one study that combines cross-section and time-series data, theestimated realizations response is so large that lower tax rates oncapital gains increase revenue even if tax rates fall below 20 percent.Other cross-section studies have found much smaller responses. Therange of results from time-series studies is narrower; the results ofsome of the studies imply that lowering tax rates on capital gains fromthe high level they reached in the mid-1970s increased revenue, butnone of them implies that the increase in the top rate from 20 percentto 28 percent in the 1986 act caused a revenue loss.

The statistical estimates in this paper extend previous time-serieswork to cover the period through 1985. The paper also develops newestimates of average marginal tax rates on capital gains, andexamines the realizations of different subgroups of the population.

THE RECENT HISTORY OF CAPITAL GAINSTAXES AND REALIZATIONS

Realizations of long-term capital gains (in excess of net short-termlosses) have generally increased with the growth in the economy,rising from $7 billion in 1954 to $165.5 billion in 1985. The growth inrealizations was especially rapid in the 1960s and after 1978. Theratio of realized long-term gains to gross national product (GNP) rose

xiv HOW CAPITAL GAINS TAX RATES AFFECT REVENUES March 1988

from less than 2 percent in 1954 to a peak of 4 percent in 1968,declined to 2 percent in 1975, and then increased sharply after 1978 to4 percent in 1985. Revenue from capital gains taxes between 1954 and1983 ranged from 0.3 percent to 0.6 percent of GNP, and from 3 per-cent to 6 percent of individual income tax revenues. Recently, theshare of revenues attributable to capital gains has risen sharply from4.3 percent in 1982 to 7.3 percent in 1985.

Realizations of long-term capital gains are highly concentratedamong the top income groups. In 1984, taxpayers with adjusted grossincome (AGI) in excess of $200,000 accounted for over 42 percent ofrealized gains; taxpayers with income over $100,000 accounted for 54percent of gains. The share of gains realized by upper-income groupsrose when gains were growing rapidly and declined when gains werestable or falling. For example, the top 1 percent of returns ranked byAGI accounted for 50 percent of realized long-term gains in 1968, only33 percent between 1975 and 1978, and about 55 percent between1982 and 1985.

It is convenient to divide capital gains taxation over the1954-1985 interval into three distinct periods: 1954-1969, 1969-1978,and 1979-1985. In the first period, capital gains tax rates were lowand stable. Taxpayers were allowed to deduct 50 percent of long-termgains from taxable income. In addition, long-term gains were allowedan alternative tax rate of 25 percent. In the second period, beginningwith the Tax Reform Act of 1969 and culminating in the Tax ReformAct of 1976, the Congress restricted the alternative tax and enactedseveral provisions that reduced the benefits of the 50 percent deduc-tion for capital gains. In the third period (1978-1985), capital gainstaxes were substantially reduced. The Revenue Act of 1978 increasedthe capital gains deduction to 60 percent and removed limits on theuse of the deduction. These changes lowered the maximum tax rate onlong-term gains from 49 percent to 28 percent. The Economic Recov-ery Tax Act of 1981 further lowered the top rate on long-term gains to20 percent.

In general, periods of low capital gains taxation have beenassociated with high levels of realizations, relative to GNP, and peri-ods of high capital gains taxation with relatively low levels ofrealizations. Other variables have also been correlated with levels ofrealizations, however, most notably the level of corporate equity

SUMMARY

values. Multiple regression analysis can be used to estimate the sepa-rate effects of different factors on realizations of capital gains.

STATISTICAL EVIDENCE ONCAPITAL GAINS AND TAXES

Taxpayers may wish to sell income-producing assets and realizecapital gains either to rearrange their financial portfolios or to financeadditional current consumption or investment in consumer durables(such as houses). The total amount of gains that can be realized forthese purposes varies positively with the available stock of accruedgains in taxpayers' portfolios. Realized gains for consumption and in-vestment in consumer durables are likely to vary positively with thelevel of total economic activity.

The stock of accrued gains on assets subject to the capital gainstax is equal to the difference between total wealth in such assets andtheir tax basis—that is, their cost to the owners. Accrued gains cannotbe observed directly because there are no data on the tax basis ofassets. As a proxy for the stock of accrued gains, the study uses mea-sures of wealth.

The investment and consumption motives mentioned above arerepresented in the study's equations by including the value of cor-porate equities held by individuals, GNP, the price level, and mea-sures of the marginal tax rate on realized gains; these variables areused to explain the annual level of realizations of net long-term capi-tal gains. Some of the equations include real values of GNP and cor-porate equities and the price level as separate variables explainingrealizations in current dollars; others use real levels of equity andGNP as independent variables to explain realizations of gains in con-stant dollars. Some equations also test for a cyclical effect by in-cluding the rate of change of GNP as well as its level.

Realized gains are found to be positively related to the value ofcorporate equity holdings, GNP, and the rate of change in GNP, andnegatively related to marginal tax rates on gains. The coefficients onthe tax rate variable (the realizations response) imply that a one-percentage-point increase in the marginal tax rate on gains (for

xvi HOW CAPITAL GAINS TAX RATES AFFECT REVENUES March 1988

example, from 19 percent to 20 percent or from 29 percent to 30percent) reduces realizations of long-term gains by between 3.1 per-cent and 3.9 percent. These point estimates imply that revenue fromcapital gains taxes would be maximized at marginal tax rates between25 percent and 33 percent.

The same estimates were made separately for the top 1 percentand bottom 99 percent of taxpayers, ranked by adjusted gross income.The realizations response estimated for the top 1 percent was almostidentical to that for the entire population. The estimate for the bottom99 percent, however, was much less precise: the realizations responsefor the bottom 99 percent was not significantly different either fromzero or from the estimated response for the top 1 percent.

The estimated responses represent both the long-run andshort-run effects of changes in tax rates on realizations, if tax rates inprevious years are assumed to have no influence on the current levelof realizations. Experiments with lagged tax rate terms failed to re-veal a consistent and statistically significant relationship betweenprevious years' tax rates and current gains. But the data may not beadequate to permit identification of complex timing relationships thatdo actually exist. There are theoretical reasons to suspect that the im-mediate effect of tax rate changes is greater than the permanenteffect. If so, the estimates reported in this paper overstate the long-run effect of tax rate changes on the realizations of gains.

Aside from the timing issue, the estimates of the realizationsresponse are subject to two types of uncertainties. First, the estimatesare quite sensitive to changes in the other variables used to explainrealizations. Second, even the estimates with a limited number ofsimple equations reveal a significant standard error in the estimate ofthe tax rate coefficient—that is, a change in the marginal tax ratecould have a wide range of possible responses. A 95 percent confidenceinterval around the estimates of the realizations response for theentire population shows that a one-percentage-point increase in themarginal tax rate could reduce capital gains realizations by as little as0.5 percent and by as much as 5.9 percent. The low response impliesthat any increase in marginal tax rates up to 100 percent would in-crease revenue, while the high response implies that revenue would bemaximized at a rate of only 17 percent.

SUMMARY xvii

Simulations of individual tax payments using the most likelyestimates of the realizations response all find that the 1986 act willincrease revenue from capital gains taxes in the long run, and thatlowering the top rate on gains to 15 percent would reduce revenue.The estimated realizations responses from four alternative equationsimply that the 1986 act will lead to an annual increase in revenuefrom capital gains taxes of between $2.6 billion and $5.9 billioncompared with previous law; this amount is much less than the in-crease that would occur if taxpayers were assumed to have no behav-ioral response, in which case the revenue pickup would be $22.4 bil-lion. Simulations with a 15 percent maximum rate on capital gainsshow an annual revenue loss of between $3.9 billion and $7.8 billion,compared with current law.

EVALUATION OF FINDINGS

The new statistical results in this study are consistent with previousstudies, which found that higher marginal tax rates on capital gainsreduced realizations. These results are also broadly consistent withofficial estimates of the revenue effect of the 1986 act, which showed arevenue pickup from raising capital gains taxes that was much lowerthan if realizations had been assumed to be fixed. On the other hand,the realizations response estimated in this study is much smaller thanthe response estimated in some studies concluding that lower tax rateson capital gains would raise revenue. The statistical estimates aresufficiently imprecise, however, that one cannot reject the possibilityof such a large response, even though it is not the most likely outcome.

The simulations do not provide any information on how long itmay take to adjust to the new revenue levels. In particular, the studydoes not consider the shifts in realizations between adjacent years thatmight occur in response to a delay in the effective date of the new law.

Nor does the study consider other behavioral responses that arepotentially important. In particular, restoring the preferential treat-ment of capital gains might reduce other forms of taxable income. Forexample, it could lower dividend and interest payments by increasingthe relative tax advantage of retained earnings, and could raisedepreciation deductions by increasing the turnover of real estate. The

xviii HOW CAPITAL GAINS TAX RATES AFFECT REVENUES March 1988

revenue consequences of such responses have not been estimated inthis or any other study; but if they occurred, the revenue loss fromlowering the capital gains tax would be greater than- indicated by theestimates in this study.

CHAPTER I

INTRODUCTION

In the Tax Reform Act of 1986, the Congress eliminated the 60 percentdeduction for long-term capital gains. The intent was to raise reve-nue to help finance the reduction in tax rates on ordinary income, topreserve distributional neutrality between higher- and lower-incometaxpayers, and to simplify the tax system by largely removing distinc-tions between capital gains and ordinary income and between short-term and long-term capital gains. The elimination of the capital gainsdeduction more than offset the drop in marginal tax rates, leaving taxrates on realized capital gains higher than under previous law.

The amount of revenue raised by the higher tax rates on capitalgains is uncertain because it depends on how realizations will respondto the higher tax rates. If realizations of gains fall substantially, littlenet revenue may be generated; realizations could even fall so far thatrevenue is lost under the higher rates.

The uncertainty about realizations exists because taxpayers haveconsiderable discretion over whether and when to pay capital gainstaxes. Capital gains are taxed only when the gains are realized bysale or exchange of assets, and gains on assets transferred at death arenever taxed. When capital gains tax rates are raised, taxpayers maydecide to defer taking gains or even to hold onto assets for as long asthey live, passing the accrued gains to their beneficiaries tax free.They may perceive the cost of being "locked in" to existing assets aslower than the tax consequences of selling.

A number of economists have used econometric models to measurehow much people alter their realizations of gains in response to taxrate changes. More recently, two studies have used the conclusions ofsome of the earlier research to simulate the effects of the 1986 act onrevenue from capital gains taxes, and have asserted that mostacademic research supports a conclusion that the capital gains

2 HOW CAPITAL GAINS TAX RATES AFFECT REVENUES March 1988

provisions in the 1986 act would reduce revenue.!,/ It has also beensuggested that lowering the capital gains tax rate to 15 percent wouldincrease revenue.2/ These conclusions have been challenged by otheranalysts, who have questioned both the methodology used in esti-mating the taxpayers' response and the application of those results inthe revenue simulations.3/

This study reviews the results of previous studies and presentsnew evidence on the relationship between capital gains realizationsand tax rates, based on recent historical data. Statistical estimates ofthis relationship are presented for the entire taxpaying populationtogether, and for the top 1 percent and bottom 99 percent of tax re-turns separately .47 The new estimates of the realizations response areused to simulate the revenue consequences of the capital gains provi-sions in the 1986 act, and to estimate the consequences of reducing themaximum tax rate on capital gains to 15 percent.

The revenue comparisons among different tax laws are meant toillustrate differences that might be expected if either current law, pre-1986 law, or post-1986 law with a 15 percent maximum rate on capitalgains had been permanently in effect, and do not reflect the effects ofshort-run adjustment patterns. Therefore, they are not directly com-parable to estimates of the five-year effects of the 1986 act or of newproposed legislation that have been prepared as part of the legislativeor budget process. What they do illustrate is the steady-state conse-quences of changing tax rates on capital gains, and the relationshipbetween revenue estimates that assume no behavioral response andestimates that assume the response would be similar to estimated re-sponses to past changes in the tax law.

1. See Lawrence B. Lindsey, "Capital Gains Taxes Under the Tax Reform Act of 1986: RevenueEstimates Under Various Assumptions," National Tax Journal (September 1987); and PeatMarwick, "The Revenue Effect of the Increase in the Capital Gains Tax Rate Enacted in the TaxReform Act of 1986," paper submitted to American Council for Capital Formation, Center for PolicyResearch (June 1987).

2. Statement of Mark A. Bloomfield, President of the American Council for Capital Formation, beforethe Committee on Ways and Means, U.S. House of Representatives, July 8,1987.

3. Eric W. Cook and John F. O'Hare, "Issues Relating to the Taxation of Capital Gains," National TaxJournal (September 1987), and Jane G. Gravelle, "A Proposal for Raising Revenue by ReducingCapital Gains Taxes?" Congressional Research Service, June 30,1987.

4. The top 1 percent of tax returns ranked by adjusted gross income (AGI) account for about half ofrealized capital gains.

CHAPTER I INTRODUCTION 3

The new results in this paper are not meant to be definitive orfinal estimates of the response of capital gains realizations andrevenue to changes in tax rates on capital gains. As will be shown, theestimates of the tax effect vary depending on how the tax rates andother variables are represented. Because changes in rates and realiza-tions have offsetting effects on revenue, fairly modest uncertaintyabout the degree of responsiveness of realizations to tax rate changescauses significant uncertainty about revenue effects. In addition, allthe econometric studies of capital gains realizations, including thisone, must confront serious methodological difficulties. For these rea-sons, revenue estimators must necessarily supplement conflictingstatistical evidence with judgment about how markets are likely towork. Finally, changes in tax rates on capital gains may affect notonly realizations, but also other components of the income tax base.For example, if higher capital gains taxes cause companies to pay outa larger share of their profits as taxable dividends, income tax reve-nues may increase even if revenues from the sale of assets decline.

CHAPTER II

PREVIOUS RESEARCH ON THE REVENUE

EFFECT OF CAPITAL GAINS TAXES

A number of earlier studies have estimated how much taxpayerschange their capital gains realizations in response to a change in thetax rate on those realizations. This behavioral response is consideredthe key to the revenue effect. If taxpayers change their realizationslittle in response to a tax rate change, then an increase in tax rateswill raise revenue and a decrease will lose revenue. If taxpayerschange their realizations by a large enough amount, the revenueeffects will be reversed; an increase in tax rates will then lose revenueand a decrease will raise revenue.

The summary of studies provided here shows a wide range ofestimated responses and implied revenue effects. While the studiesconsistently find that taxes discourage realizations, they disagree asto whether the discouragement resulting from a tax rate increase islarge enough to offset the revenue effect. No consensus has emergedas to whether the tax changes enacted in recent years have causedrevenues to move in the same direction as the rate changes or in theopposite direction. The studies summarized below, their main fea-tures, and their revenue findings are listed in Table 1 at the end of thischapter.

APPROACHES TO ESTIMATINGTHE REALIZATIONS RESPONSE

Studies that estimate the effect of tax rates on capital gainsrealizations use two different approaches: the cross section and thetime series. Cross-section studies compare the behavior of differenttaxpayers or taxpayer groups in the same year or over several years.They attempt to explain differences in capital gains realizationsamong taxpayers by differences in marginal tax rates, while control-ling for the effects of other characteristics of taxpayers, such as totalincome, age, family status, and dividends (used as an indicator of stock


ownership). These estimates, based on differences among taxpayers,are then used to infer what might happen to the realizations of all tax-payers if the tax law was changed.

Time-series studies, in contrast, attempt to observe directly the ef-fects of differences in marginal tax rates over time on differences inaggregate capital gains realizations. These studies also control forother variables believed to influence realized gains, such as realincome, wealth, and the price level. Some studies attempt to estimatea lag structure-that is, the time pattern of response of realizations tochanges in tax rates. Because marginal tax rates differ among tax-payers, all the time-series analyses either use the top rate on capitalgains as a proxy for all capital gains rates or construct a marginal taxrate series that represents a weighted average of the rates faced bytaxpayers who realize capital gains.

Major methodological problems arise in both approaches. Cross-section studies are able to use a tremendous amount of detailed dataon individual taxpayers, but confront a major difficulty in that thegroup of taxpayers under observation are all facing the same tax law.Consequently, all the differences in tax rates among the units beingobserved reflect some characteristics of the taxpayers themselves; it isnever entirely clear whether the differences in capital gains realiza-tions reflect differences in taxpayer characteristics or are the indepen-dent effects of tax rate differences. Other problems in cross-sectionstudies are the commingling of transitory and permanent changes intax rates, and the limitations in the data provided on tax returns.

Time-series studies overcome the most serious problem in thecross-section studies because they make it possible to identify andmeasure changes in tax rates over the sample period that are indepen-dent of taxpayer behavior. The major problem in the time-seriesstudies is that the number of observations is quite limited. As aresult, the estimated relationship between realizations and tax ratesis highly dependent on the other factors that are hypothesized to influ-ence capital gains realizations. The limitations in data also make itdifficult to identify the time pattern of responses to changes in capitalgains taxes. Finally, some biases may result from the need to useaggregate data that combine taxpayers who may have very differentbehavioral responses. A more complete discussion of methodological

CHAPTER II PREVIOUS RESEARCH ON THE REVENUE EFFECT 7

problems in studies of the relationship between realizations and taxrates is provided in Appendix B.

MEASURES OF REVENUE EFFECTS

The studies examined here employ three methods of drawing revenueimplications from their estimates of the realization response. In onemethod, the estimated response is applied to a population of taxpayersto simulate directly the revenue consequence of a specific tax change.In the other two, the revenue effects of tax changes are summarized bymeasures of "tax rate elasticities" or "revenue-maximizing tax rates."

The measure of tax rate elasticity is the percent change in reali-zations estimated to result from a 1 percent change in tax rates. If a 1percent increase in tax rates is estimated to cause a 1 percent declinein realizations, then the elasticity is minus one. Revenues would beunchanged because the percentage increase in rates would be justoffset by an equal percentage decline in realizations. If the percentagedecline in realizations is estimated at less than 1 percent, the absolutevalue of the elasticity is less than one, and a tax increase would in-crease revenues.l/ On the other hand, if the percentage decline inrealizations is estimated to be greater than 1 percent, the elasticity isgreater than one (for example, -1.5 or -3.0), and a tax increase woulddecrease revenues. Similarly, with a decrease in tax rates, if the ab-solute value of the elasticity is estimated to be between zero and one,revenues would also decrease; and if the elasticity is estimated to begreater than one, revenues would increase.

While elasticity is a convenient summary, it is not a guide to allsituations. Elasticity itself can change with tax rates, with the in-comes of taxpayers, and with the mix of assets being realized. Anelasticity estimated to be between zero and one at the average tax ratestudied might be greater than one at the highest tax rate. Similarly,an elasticity estimated for the highest-income taxpayers might be in-

Since the elasticity is a negative number, a lower value implies a higher absolute value. That is, ifthe elasticity is -0.5, it is greater than -1, but its absolute value is less than one. In the followingdiscussion, the use of the term elasticity refers to its absolute value.


appropriate for all taxpayers together, and an elasticity estimated forcorporate stocks might be different from that for real estate.

A few studies summarize their revenue effects by calculating arevenue-maximizing tax rate instead of an elasticity. The concept of arevenue-maximizing rate is based on the assumption that theelasticity of realizations with respect to tax rates increases as ratesrise. At very low tax rates, the tax may be viewed as inconsequentialso that a large percentage tax increase has little effect on realizationsand revenues rise (an elasticity between zero and one). At higher taxrates, the tax becomes more consequential and realizations are moreaffected. At high enough rates, the realizations may be sufficientlyaffected that any further increase in tax rates causes realizations todecline by a larger percentage than the rate increase (elasticitygreater than one). At this point, revenues fall for any further tax rateincrease, and thus a revenue-maximizing rate has been reached.

THE STUDIES

The eight studies summarized below differ widely in their attempts toestimate how realizations respond to tax rates. Three use cross-sec-tion analysis, three use time-series analysis, one uses both separately,and one combines cross-section and time-series. Some cross-sectionstudies consider only corporate stock sales by high-income taxpayers;others include all taxable asset sales and a representative sample ofall taxpayers. Time-series studies, and the combined time-series andcross-section study, differ most in the way they measure average taxrates and the pool of outstanding gains available to be realized.

Feldstein, Slemrod, and Yitzhaki

These authors (FSY) provided one of the first thorough studies of howcapital gains realizations respond to changes in tax rates.2/ Theyapplied cross-section analysis to a special sample of 1973 tax returns,

2. Martin S. Feldstein, Joel B. Slemrod, and Shlomo Yitzhaki, "The Effects of Taxation on the Sellingof Corporate Stock and the Realization of Capital Gains," Quarterly Journal of Economics, vol. 94(June 1980), pp. 777-791.


known as the Capital Assets Study. This data file gives extensive in-formation on asset sales and oversamples high-income returns whereasset sales are concentrated. For most of their analysis, FSY limitedtheir sample to tax returns reporting $3,000 or more in dividends.

FSY attempted to explain only realizations of gains on corporatestock. Gains on stock were explained by tax rates on realized gains,dividends, other income, and whether a taxpayer was over 65 years ofage.3/ FSY found realized gains on stocks to be highly sensitive to taxrates for those with over $3,000 of dividends, but not for other tax-payers.4/

For those with over $3,000 in dividends, FSY simulated therevenue effects of two tax changes. One would have capped the rate oncapital gains at 25 percent (in 1973, top capital gains rates ranged upto over 45 percent for some taxpayers—see Chapter HI). This tax re-duction was found to raise revenues. The second proposal would haveraised rates on all capital gains by eliminating the exclusion for long-term gains. This tax increase was found to lose revenue. Thus, FSYfound revenues moving in the opposite direction of tax rate changes intwo cases of fairly large changes in tax rates.

Minarik

Minarik estimated capital gains equations using the same data baseas FSY but a different estimating methodology, and reported very

3. The marginal tax rate paid on capital gains depends both on the legislated rate structure and onthe level of gains a taxpayer chooses to realize. Because the studies seek to estimate the effect ofrates on realizations, they take steps to purge their tax rate measure of the effect of differences inrealizations. FSY used a predicted rate based on the taxpayer's marginal rate on the first dollar ofrealizations, and on the last-dollar marginal rate if the taxpayer's realized gains were equal to theaverage for similar taxpayers.

4. An earlier study by Feldstein and Yitzhaki had suggested the possibility of a negative relationshipbetween realized gains and tax rates by showing that total stock sales were discouraged by highertax rates and encouraged by lower tax rates. See Feldstein, Slemrod, and Yitzhaki, "The Effects ofTaxation," pp. 781-786, and Martin S. Feldstein and Shlomo Yitzhaki, "The Effect of the CapitalGains Tax on the Selling and Switching of Common Stock," Journal of Public Economics, vol. 9(February 1978), pp. 17-36.


different findings.5/ For those with $3,000 or more of dividends,Minarik reported elasticities of-0.44 in one specification and -0.79 inanother. These estimates suggest that revenues move in the samedirection as a tax rate change on capital gains. Minarik simulated theeffect of the rate reductions in the 1978 tax act and found them losingrevenue.

In a subsequent paper, Minarik showed that the main reason forthe difference between his and FSY's results was that FSY did notadjust for the oversampling of high-income taxpayers in the CapitalAssets Study.6/ Minarik argued that this omission biased FSY's esti-mated response upward and showed that use of a weighted regressiontechnique produces a much smaller response. FSY replied that theirhigher estimate resulted from the higher-income taxpayers who areoverrepresented in the sample being more responsive to tax rates thanothers, and were thus able to argue that their estimate is appropriatefor evaluating tax changes that primarily affect the very highest-income taxpayers.7/

One problem common to both the FSY and the Minarik studieswas that the data base did not permit them to separate the taxresponse into temporary and permanent components. Taxpayers fac-ing temporarily low tax rates in a given year because of, for example,above-average business losses, could take advantage of those rates bybunching capital gains realizations in that year. Those taxpayerswould not necessarily continue the same level of realizations if capitalgains taxes were permanently lowered to this rate. Using only oneyear's data, neither FSY nor Minarik could distinguish between atransitory response to temporarily lower rates and a response to per-manently lower ones, and therefore the estimates in both papers mayhave overstated the permanent response to a tax change.

5. Joseph J. Minarik, "Capital Gains," in Henry J. Aaron and Joseph A. Pechman, eds., How TaxesAffect Economic Behavior (Washington, D.C.: Brookings Institution, 1981).

6. Joseph J. Minarik, "The Effects of Taxation on the Selling of Corporate Stock and the Realization ofCapital Gains: Comment," Quarterly Journal of Economics, vol. 99 (February 1984), pp. 93-110.

7. Martin S. Feldstein, Joel B. Slemrod, and Shlomo Yitzhaki, "The Effects of Taxation on the Sellingof Corporate Stock and the Realization of Capital Gains," Quarterly Journal of Economics, vol. 99(February 1984), pp. 114-117;Minarik, "Capital Gains," pp.255-265.


Auten and Clotfelter

Auten and Clotfelter were able to separate permanent and temporarytax rate responses by using the Treasury Department's Seven-YearPanel of Taxpayers.8/ This data base included all tax returns filed bya sample of the same taxpayers for the years 1967 through 1973. Thepermanent tax rate was measured as the average of marginal ratesover the current and the preceding two years, and the transitory com-ponent was the difference between the current and the permanentrate. With this separation, Auten and Clotfelter found a large transi-tory response and a smaller permanent response. The elasticity fortransitory tax changes was greater than one in most specifications,while the elasticity for permanent changes was less than one. Per-manent elasticities of-0.37 and -0.55 were reported in two representa-tive equations. These elasticities indicate that revenues would movein the same direction as tax changes, at least for small changes fromrates prevailing during the sample period.

The finding of a significant transitory component suggests thatthe response estimated by FSY probably overstates the permanentresponse to a tax change. However, Auten and Clotfelter point outthat differences between the Capital Assets File and the Seven-YearPanel may also account for differences between their permanentresponse and the higher response estimated by FSY. First, the Seven-Year Panel combined all realizations together, so the lower perma-nent response could be partly the result of a lower responsiveness inthe sale of assets other than corporate stocks. Second, the panel hadtoo few high-income taxpayers to obtain separate estimates for thosewith dividends as high as in the earlier studies. Thus, the lower per-manent response may also reflect less responsiveness among middle-and lower-income taxpayers, which the previous studies and Autenand Clotfelter all found to be the case.9/

8. Gerald E. Auten and Charles T. Clotfelter, "Permanent Versus Transitory Tax Effects and theRealization of Capital Gains," Quarterly Journal of Economics, vol. 97 (November 1982), pp. 613-632.

9. Auten and Clotfelter did find limited evidence that higher-income taxpayers are more responsive,but the income classes were all lower than in the earlier studies. Auten and Clotfelter, "PermanentVersus Transitory Tax Effects," p. 629 and footnote 26.


Auten

Auten used time-series data to explain total realized gains of alltaxpayers from 1951 to 1980 by the tax rate on capital gains, a con-structed measure of unrealized gains, and total AGI other than fromcapital gains.10/ The measure of the tax rate used was an averagemarginal capital gains tax rate for taxpayers with over $50,000 ofAGI. These taxpayers represent 10 percent or less of all taxpayers butaccount for a large share of capital gains realizations.ll/ In additionto the direct effect of taxes on realizations, Auten included an effect oftaxes on asset prices that indirectly affected realizations.12/

Auten found that taxes had statistically significant direct andindirect effects on realizations. About 10 percent of the response ofrealizations to tax rate changes came through the indirect effect oftaxes on asset prices. His simulations showed the tax reductions inthe 1978 act gaining revenue but those in the 1981 act losing revenue.

The Treasury Department

As part of a comprehensive study of the effects of the 1978 capitalgains tax changes, the Treasury Department analyzed revenue effectsof capital gains taxes using both cross-section and time-seriesmodels. 137

Treasury's cross-section analysis used a 1971 through 1975 panelof taxpayers and separated out temporary and permanent taxchanges. The study found a substantial transitory response, but evenso, the permanent elasticity was greater than one. Depending on the

10. Gerald E. Auten, "Capital Gains: An Evaluation of the 1978 and 1981 Tax Cuts," in Charls E.Walker and Mark A. Bloomfield, eds., New Directions in Federal Tax Policy for the 1980s(Cambridge, Mass.: Ballinger, 1983). The available time-series data on realizations cannot bedisaggregated by asset type, so differences in response for corporate stocks, real estate, and otherassets have not been explored by Auten or others through time-series analysis.

11. As in other studies, the tax rate was adjusted to avoid reverse influences of realizations on theaverage tax rate.

12. Auten found that, for example, lower taxes increased the value of corporate stock, which in turnincreased realized gains. See Auten, "Capital Gains," p. 136.

13. Office of the Secretary of the Treasury, Office of Tax Analysis, Report to Congress on the CapitalGains Tax Reductions of 1978 (September 1985), pp. 157-185.


specification, the elasticity for aggregate realizations ranged from-1.16 to -2.20 at taxpayers' marginal tax rates in those years. (Themaximum rate on net long-term gains in 1971-1973 ranged from 38percent to 45.5 percent. See Chapter HI.) As all of the estimatedelasticities are greater than one, they suggest that revenues wouldmove in the opposite direction of tax rate changes. Treasury explicitlysimulated the revenue effects of the capital gains tax reductions in the1978 and 1981 acts and found both raised revenue.14/ Realizations ofgains on corporate stock were found to be more responsive to tax ratesthan realizations of other assets. The elasticity for realizations ofcorporate stock was estimated at -2.07 compared with -0.71 for realestate and -0.43 for all other assets.

Treasury's time-series analysis extended Auten's initial study tomore recent years, 1954 through 1982, and made several changes.Tax rates were calculated for taxpayers with real incomes of $200,000or more instead of those with dollar incomes over $50,000. Tax ratesin the previous year were also included in case the first-year responseto tax changes differed from the longer-term response. Instead of acomplex construction of outstanding gains on a variety of assets,Treasury approximated all such gains with two variables—the value ofcorporate stock and personal income. The value of corporate stockmeasured gains on a major component of all realized gains, and per-sonal income served as a proxy variable for gains that generally growwith the size of the economy. Inflationary increases in stock valuesand personal income were allowed to have separate effects on realiza-tions by using real changes in these variables and including a priceindex. The Treasury model explained changes in realizations fromyear to year instead of the level in each year.

Treasury's time-series estimates of the effects of the 1978 and1981 acts were qualitatively similar to the earlier findings of Auten,but differed from Treasury's cross-section results. Simulations foundthe rate reductions of the 1978 act gaining revenue, but those of the1981 act losing revenue. These effects held whether or not an indirecteffect of taxes on asset values was included. Treasury also found theresponse in the first year after a tax change to be larger than the long-run response.

14. The simulations applied the behavioral responses estimated in the cross-section model to a sampleof tax returns for 1979.


Congressional Budget Office

As part of a larger study evaluating the distributional and supply-sideeffects of the 1981 act, the Congressional Budget Office (CBO)modified the Treasury time-series analysis and extended it through1983.15/ A constructed average tax rate on gains for all incomeclasses was used for the first time and, as in Auten's study, annualrealizations were explained instead of changes from year to year. Theelasticity in the first year of a tax change can be calculated to be -0.56and the elasticity in the second and later years is -0.23. These esti-mated elasticities suggest that revenues will move in the same direc-tion as tax rates, at least for changes in rates around the average ofthe post-Korean War experience. The CBO study did not, however,draw any direct conclusions about the revenue effects of changingcapital gains taxes. The difference in response between the first andlater years was not statistically significant as it had been in the Trea-sury time-series model.

Cook and O'Hare

Cook and O'Hare also used time-series data through 1983.167 Theirfocus was on tax-induced changes in holdings of gains-producing andincome-producing assets. Thus, they included the spread betweenordinary income tax rates and capital gains tax rates, as well as thelevel of capital gains rates, in separate equations explaining realizedgains and the sum of interest and dividend income. The tax rate oncapital gains and the spread were measured at the highest rates ineffect for both types of income. As with other studies, the level of taxrates had a negative and statistically significant effect on realizations.In addition, a greater spread was found to increase capital gainsrealizations, but the effect was not statistically different from zero.Cook and O'Hare did not report elasticities or simulate the revenueeffects of their estimates. They did find evidence that an increase in

15. Congressional Budget Office, "Effects of the 1981 Tax Act on the Distribution of Income and TaxesPaid," Staff Working Paper (August 1986). The estimates were used to calculate the portion of th«rise in reported income of the top 1 percent of taxpayers between 1980 and 1983 that might beattributable to lower capital gains tax rates.

16. Eric W. Cook and John F. O'Hare, "Issues Relating to the Taxation of Capital Gains," National TaxJournal, vol. 40 (September 1987), pp. 473-488.


the spread between ordinary income tax rates and capital gains taxrates could reduce interest and dividend income. This is one channelby which a lower capital gains tax rate could reduce the tax baseinstead of increasing it, as discussed below.

Lindsey

Lindsey combined cross-section and time-series analysis by dividingtaxpayers in each year from 1965 through 1982 into six AGIclasses.1/7/ Thus, the comparisons within a year are among classes oftaxpayers rather than among individuals as in other cross-sectionstudies. Lindsey explains differences in realizations by differences intax rates and constructed measures of "tradable wealth," "nontrad-able wealth," and recent appreciation in tradable wealth. Tradablewealth includes those assets likely to yield capital gains: corporatestock, rental and owner-occupied real estate, and equity in non-corporate businesses.

The study finds realizations to be very sensitive to tax rates. Theestimated equations use two functional forms that constrain elastici-ties to rise with tax rates and facilitate calculation of a revenue-maximizing tax rate. The revenue-maximizing tax rate is found to bebetween 14 percent and 20 percent. A subsequent simulation byLindsey using these estimates finds that the 1986 act, which raisedmaximum rates from 20 percent to 28 percent, loses revenue.18/

Overview

The studies of capital gains realizations and tax rates described abovereach different conclusions about the revenue effect of changingcapital gains taxes. No consensus exists that a change in tax rateswill cause revenues to move in the same or in the opposite direction ofthe tax change. (See Table 1, at the end of the chapter, for a summaryof the main characteristics and findings of each study.)

17. Lawrence B. Lindsey, Capital Gains: Rates, Realizations and Revenues, Working Paper No. 1893(National Bureau of Economic Research, Inc., April 1986).

18. Lawrence B. Lindsey, "Capital Gains Taxes Under the Tax Reform Act of 1986: Revenue EstimatesUnder Various Assumptions," pp. 495-503.

82-667 0 - 8 8 - 2


Differences are much wider among the cross-section studies thanamong the time-series studies. FSY and the Treasury cross-sectionfind realizations so sensitive to tax rate changes that revenues movein the opposite direction of the rate changes. The studies by Minarikand by Auten and Clotfelter find realizations less sensitive to taxrates. Minarik's simulation found a tax rate reduction losing revenue,and Auten and Clotfelter's reported elasticities also suggest revenueswould move in the same direction as tax rate changes.

The time-series studies show more consistency, possibly becausethey all use much the same data. Auten and the Treasury time-seriesstudy agree that reductions in tax rates from their high levels beforethe 1978 act gained revenue, but that the further reductions in the1981 act lost revenue. The elasticities in the CBO study of the 1981act suggest that revenues would move in the same direction as taxrates, at least for tax changes around the historical rate structure.

The combined cross-section and time-series results of Lindsey arequalitatively similar to the cross-section results of FSY and Treasury.Realizations are very sensitive to tax rates, and Lindsey's simulationsfind that the tax rate increase in the 1986 act loses revenue.

In spite of their differences about revenue effects, the studies havesome general areas of agreement. Almost all find that tax rate in-creases discourage realizations. Three of the four cross-section studiestest for and find a greater responsiveness to tax rates among higher-income or wealthier taxpayers than among the population in general.The Treasury cross-section study finds that realizations of corporatestock are more sensitive to tax rates than realizations of other assets;this finding is consistent with the high responsiveness for stock salesfound by FSY. Two time-series studies suggest that realizations mayrespond more in the first year of a tax change than in later years.

OTHER REVENUE EFFECTS OFCAPITAL GAINS TAX CHANGES

Changes in capital gains tax rates can affect revenues more broadlythan through the response of capital gains realizations. They maycause components of the tax base, or even the size of the economy, to


change. Changes in the tax base caused by changes in capital gainsrates could well affect revenues, but induced changes in the size of theeconomy are not likely to be large enough to affect revenues much, atleast in the near term. These broader revenue effects are not incor-porated in the empirical findings summarized above, although theyare discussed in the Treasury study, the study by Cook and O'Hare,and Gravelle's review of Lindsey's research. 19/

Other types of property income are the components of the tax basethat are most likely to be affected by changes in capital gains taxrates. For example, an increase in capital gains tax rates that dis-courages realizations could also encourage corporations to increasedividend payouts and rely more on debt financing than retainedearnings. Higher dividend and interest payments would raise individ-ual income tax revenues. More generally, any substitution of ordinaryincome for capital gains would increase revenue as long as the tax rateon ordinary income was greater than the tax rate on capital gainsbefore the increase. Cook and O'Hare found limited evidence of theseother changes in the tax base.20/ They found that the spread betweentax rates on ordinary income and on long-term capital gains signifi-cantly influences the level of dividend and interest income received bytaxpayers.

The overall effect on the tax base of changes in capital gains taxrates has not, however, been estimated in any systematic way. A morecomplete discussion of the types of revenue feedbacks that might beexpected is provided in Chapter IV.

19. Jane G. Gravelle, "A Proposal for Raising Revenue by Reducing Capital Gains Taxes?".

20. Cook and O'Hare, "Issues Relating to the Taxation of Capital Gains," pp. 485-487.


TABLE 1. SUMMARY OF PREVIOUS STUDIES ON REVENUEEFFECTS OF CAPITAL GAINS TAXES

Authors

Feldstein, Slem-rod, Yitzhaki

Minarik

Auten andClotfelter

Auten

Data

CROSS SECTIONof individual tax-payers in 1973

Includes tax-payers withover $3,000of dividends

Limited torealizationson stocks

CROSS SECTION,same as Feld-stein, Slemrod,Yitzhaki

CROSS SECTIONon individual tax-payers, paneldata 1967-1973

Represents tax-payers with over$200 in dividendsor rents and$5,000 in AGIless gains

Includes allreported assetsales

TIME SERIES ontaxpayer totalsand aggregateeconomic data,1951-1980

SpecialFeatures

No weighting ina sample over-representing tax-payers with high-est incomes

Weighted toreflect popu-lation oftaxpayers

Includes morenontax influenceson realizationsthan other studies

Separate re-sponses totemporary andpermanent taxchanges

Tax rate is pre-dicted averagefor taxpayerswith AGI over$50,000

Revenue predic-tion includes in-direct effect ofcapital gains taxeson asset values

Size of TaxRate Response

Large response

Larger responseat higher AGI

Elasticity notreported

Elasticities of-0.44, -0.79

Higher elasticityat higher incomes

Permanentelasticities of-0.37 and -0.55

Higher elasticityat higher incomes

Temporary re-sponse largerthan permanentresponse


RevenueEffects

Reducing toprate increasesrevenue onstock sales

Removing ex-clusion losesrevenue

1978 tax reduc-tions loserevenue onstock sales

No simulations

Long-run reve-nues likely tomove with taxrate changesbecause elastic-ity betweenOand-1

1978 tax reduc-tion raisesrevenue but1981 reductionloses revenue

(Continued)

CHAPTER H PREVIOUS RESEARCH ON THE REVENUE EFFECT 19

TABLE 1. Continued

Authors

Treasury

Treasury

CBO Study of1981 Tax Act

Cook andO'Hare

Lindsey

Data

CROSS SECTIONon individual tax-payers, panel data1971-1975

Represents alltaxpayers andall asset sales




Combined CROSSSECTION andTIME SERIES onsix AGI classes oftaxpayers, 1965-1982

SpecialFeatures

Estimates re-sponses to per-manent andtemporary taxchanges, andresponses ofdifferent assettypes

Tax rate is pre-dicted averagefor taxpayerswith real AGIover $200,000

Tax rate is pre-dicted averagefor all taxpayers

Includes spreadbetween tax rateson ordinary in-come and capitalgains

Tax rate per classis unweightedaverage

Distributes aggre-gate economicdata among AGIclasses

Size of TaxRate Response

Permanent elastic-ities of -1.16 to-2.20 for all assets

-2.07 for stocks,-0.73 for real estate,-0.43 other assets


First year's re-sponse greaterthan later years'

Elasticities esti-mated as -0.56first year, -0.23 later


Spread marginallyimportant forcapital gains

Estimated coeffi-cients imply reve-nue-maximizingtax rates between14 percent and20 percent

RevenueEffects

1978 and 1981tax reductionsraise revenue

1978 tax reduc-tion raisesrevenue but1981 reductionloses revenue

No revenueeffect reported,but elasticitiessuggest revenueswould changein same direc-tion as tax rates

Not reported

1986 tax in-creases willreduce revenue

SOURCE: Congressional Budget Office.

CHAPTER III

CAPITAL GAINS TAXES AND

REALIZATIONS, 1954-1985

This chapter reviews the history of capital gains taxation and realiza-tions of capital gains for the post-Korean War period. For backgroundpurposes, the chapter begins with a brief discussion of the distributionof realized capital gains among income groups. It then presents dataon changes over time in the amount and distribution of realized capi-tal gains and revenue from capital gains taxes. These changes in capi-tal gains are compared with changes in the tax treatment of capitalgains and with changes in other economic variables.

DISTRIBUTION OF GAINS AMONG INCOME GROUPS

Realizations of capital gains are highly concentrated among the topincome groups (see Table 2). In 1984, taxpayers with adjusted grossincome (AGI) of $1 million or more (0.01 percent of all returns) ac-counted for over a fifth of net long-term gains (in excess of net short-term losses).!./ In contrast, they received only 1 percent of total AGIfrom other sources. Taxpayers with AGI of $200,000 or greater (aboutthe top quarter of 1 percent of returns) accounted for 42 percent of allgains and 4 percent of other AGI. Taxpayers with AGI of $100,000 ormore (the top percentile of returns) accounted for 54 percent of gainsand about 9 percent of other AGI.

1. To qualify as long term under current law, an asset must be held for at least one year. Before 1977,the holding period was six months. The Tax Reform Act of 1976 increased the holding period fordefining capital transactions as long term from six months to one year; the Tax Reform Act of 1984reinstated the six-month holding period for transactions between December 22,1984, and January1,1988.

In computing net gains eligible for the capital gains deduction (50 percent of net capital gainsbefore October 31, 1978, and 60 percent between October 31, 1978, and January 1, 1987), thetaxpayer would subtract from long-term gains all long-term losses and also would subtract theexcess of short-term losses over short-term gains. The gains figures in Table 2 and the other tablesin this chapter report the resulting amount of net long-term gains in excess of net short-term lossesfor taxpayers for whom this figure was positive. Net short-term gains are not shown because theyhave never been eligible for the capital gains deduction.


Realized capital gains are also significant for taxpayers with zeroor negative amounts of AGI. Taxpayers with AGI less than or equal tozero accounted for about 15 percent of all long-term realizationsamong taxpayers with AGI less than $100,000. These taxpayers areclearly not low-income people in the usual sense. They hold substan-tial assets and realize large amounts of gains per return, and thereforemust have negative AGI from other sources. The most commonsources of negative AGI are losses from proprietorships, partnerships,and farms. These losses in many cases represent the effects of taxpreferences, such as expensing and accelerated depreciation, insteadof real economic losses.

Previous research has also shown that reported capital gains aremore concentrated among the high-income groups than are taxable

TABLE 2. DISTRIBUTION OF LONG-TERM CAPITAL GAINSBY ADJUSTED GROSS INCOME (AGI): 1984

AGI Class(In thousandsof dollars)

Less than 00-55-1010-1515-2020-2525-3030-4040-5050-7575-100100 - 200200 - 500500-1,0001,000 or more

Total

SOURCE: Congressii

Share ofLong-Term

Capital Gains(In percents)

7.110.921.642.393.132.511.954.964.949.546.31

12.1512.968.19

21.30

100.00

jnal Budget Office, computed

Share ofAGI ExcludingCapital Gains(In percents)

-1.802.055.878.319.519.44

10.0018.3212.5912.944.114.472.400.721.06

100.00

L from data reported in

Share ofTax Returns(In percents)

1.0116.3316.5414.1411.558.877.68

11.146.004.681.060.770.200.030.01

100.00

Internal Revenue Service,Statistics of Income, Individual Income Tax Returns 1984.

CHAPTER IH CAPITAL GAINS TAXES AND REALIZATIONS, 1954-1985 23

interest and dividends, and that the gains were also highly concen-trated at the top in earlier years. A portion of the concentration ofgains at the top reflects the temporary movement of taxpayers intohigh-AGI groups in years in which they realize large gains. Usingpanel data, however, the Treasury showed that permanent capitalgains (defined as gains averaged over a five-year period) are alsohighly concentrated among taxpayers with high permanent AGI.2/

Finally, capital gains shown on tax forms (and reported in Table2) overstate positive real income because the tax system does not ad-just for the increase in the general price level between the time assetsare bought and the time they are sold. To compute the real gain on thesale of an asset, one must adjust the tax basis—that is, the purchaseprice—for the increase in the general price level over the period theasset was held. Previous research has revealed that, in 1973 and1977, the distribution of real gains on sales of corporate stock and non-business real estate was even more highly concentrated among thehighest-income groups than was the distribution of nominal gains.Indeed, for sales of corporate stock in 1977, total real gains were nega-tive in the aggregate and for all groups with AGI less than $100,000,but were positive for AGI groups over $100,000.37 Data that can beused to compute real gains of assets sold in years after 1977 are notpublicly available.

The extreme concentration of realized gains at the top of the in-come distribution suggests that the revenue effect of changes in thetaxation of capital gains is influenced heavily by the behavior of thehighest-income taxpayers. As discussed in Chapter IE, earlier cross-section studies have found evidence that higher-income taxpayersrespond more to tax changes than others. In particular, because the1986 Tax Reform Act raised capital gains tax rates by a much largerproportionate amount for lower- and middle-income taxpayers thanfor the highest-income groups, it is worthwhile to examine throughtime-series analysis whether the behavioral response might differ

2. Office of the Secretary of the Treasury, Office of Tax Analysis, Report to Congress on the CapitalGains Tax Reductions of 1978 (September 1985). The Treasury used panel data that followedreturns of the same taxpayers for a five-year period.

3. Ibid. A similar result for 1973 is found in Martin S. Feldstein and Joel B. Slemrod, "Inflation andthe Excess Taxation of Gains on Corporate Stock," National Tax Journal (June, 1978).


among income groups.4/ This issue is considered in the econometricwork presented in Chapter IV.

GROWTH IN CAPITAL GAINSREALIZATIONS OVER TIME

Realizations of long-term capital gains have increased greatly since1954, with very rapid growth occurring in the 1960s and between 1978and 1985. In 1985, the latest year for which data are available, long-term realizations of capital gains were $165.5 billion, compared with$7.0 billion in 1954 and $48.9 billion in 1978.

Capital gains realizations might be expected to grow in proportionto the level of total accrued gains. Total accrued capital gains cannotbe directly measured, but are likely to follow overall growth in theeconomy and in the value of corporate equities.

In general, the growth in realized capital gains has moved upwardwith the overall growth of GNP (Table 3 and Figure 1). Capital gainsgenerally increased faster than GNP in the 1950s and 1960s, reachinga peak of 4 percent of GNP in 1968. Gains declined sharply relative toGNP in the 1969-1970 recession and again following the decline of thedollar and the increase in oil prices in 1973, falling to 1.9 percent ofGNP in the recession year 1975. After 1975, capital gains began to in-crease more rapidly than GNP, with a big jump in 1979 and subse-quent large increases in 1984 and 1985. Capital gains as a share ofGNP almost doubled between 1978 and 1985, rising from 2.2 percentto 4.1 percent, an amount just above the share reached in 1968.

4. For example, for a married couple with taxable income of $30,000 in 1988, the marginal tax rate onordinary income was lowered by the 1986 act from 32 percent to 28 percent, while the marginal taxrate on long-term gains was increased from 12.8 percent to 28 percent-an increase of 219 percent.In contrast, for a married couple with income greater than $200,000, the marginal tax rate onordinary income dropped from 50 percent to 28 percent, while the rate on long-term gains increasedfrom 20 percent to 28 percent~an increase of 40 percent.

CHAPTER HI CAPITAL GAINS TAXES AND REALIZATIONS, 1954-1985 25

TABLE 3. REALIZED NET LONG-TERM GAINS COMPAREDWITH GROSS NATIONAL PRODUCT, 1954-1985

Year

19541955195619571958195919601961196219631964196519661967196819691970197119721973197419751976197719781979198019811982198319841985

Realizationsof Net Long-Term Gains(In billionsof dollars)

7.09.79.68.29.3

12.911.715.713.614.517.020.821.827.335.832.621.328.236.135.830.030.739.244.448.971.370.878.387.1

117.3135.9165.5

GrossNationalProduct

(In billionsof dollars)

372.5405.9428.1451.0456.8495.8515.3533.8574.6606.9649.8705.1772.0816.4892.7963.9

1,015.51,102.71,212.81,359.31,472.81,598.41,782.81,990.52,249.72,508.22,732.03,052.63,166.03,405.73,765.03,998.1

Ratio ofGains to GNP(In percents)

1.92.42.21.82.02.62.32.92.42.42.63.02.83.34.03.42.12.63.02.62.01.92.22.22.22.82.62.62.83.43.64.1

SOURCES: Congressional Budget Office, computed from published data of the Internal Revenue Ser-vice and the Department of Commerce, Bureau of Economic Analysis.


Figure 1.Realized Net Long-Term Gains and Gross National Product,1954-1985 (In billions of dollars)

200

150

100

50

I I

GrossNationalProduct

(Right Scale)

Long-termGains

(Left Scale)

I I I I I I I I I I I I I I I I I I I I I I

5.000

3,750

2,500

1,250

1955 1960 1965 1970

Year

1975 1980 1985

SOURCES: Congressional Budget Office, computed from published data of the Internal RevenueService and the Department of Commerce, Bureau of Economic Analysis.

Capital gains realizations have also tended to grow with the valueof corporate equities held by households (Table 4 and Figure 2).5/Household equity values have grown more rapidly than GNP since1978, which could account for some of the growth in realizations rela-tive to GNP in recent years. Realizations since 1978, however, havegrown faster than household equity values, reaching historical peaksof 8 percent to 9 percent of equity values in 1979, 1983, 1984, and1985.

5. The value of corporate equities held by households differs from an index of stock prices in tworespects. First, it reflects changes in the number of shares, as well as changes in the price pershare. Second, it reflects the long-term secular decline in the ratio of corporate equities held byhouseholds to those held by pension funds. The latter are not subject to tax.

CHAPTER in CAPITAL GAINS TAXES AND REALIZATIONS, 1954-1985 27

TABLE 4. REALIZATIONS OF NET LONG-TERM GAINS COMPAREDWITH CORPORATE EQUITY HELD BY INDIVIDUALS,1954-1985

Year

19541955195619571958195919601961196219631964196519661967196819691970197119721973197419751976197719781979198019811982198319841985

Realizations ofNet Long-

Term Gains(In billionsof dollars)

7.09.79.68.29.3

12.911.715.713.614.517.020.821.827.335.832.621.328.236.135.830.030.739.244.448.971.370.878.387.1

117.3135.9165.5

CorporateEquity

(In billionsof dollars)

235.0286.3305.1267.4373.3402.0395.5500.8437.4513.9564.7635.6575.6720.4857.9746.1728.6832.8920.6709.7494.4641.0754.1708.8721.3872.5

1,166.71,120.31,265.71,454.21,490.71,948.0

Ratio ofGains toEquity

(In percents)

3.03.43.13.12.53.23.03.13.12.83.03.33.83.84.24.42.93.43.95.06.14.85.26.36.88.26.17.06.98.19.18.5

SOURCES: Congressional Budget Office, computed from published data of the Internal Revenue Ser-vice and the Board of Governors of the Federal Reserve System.

NOTE: Mutual funds are included in the measure of corporate equity because the gains realized ontheir transactions are taxed to households, but the value of equities held for households bypension funds is excluded because the realization of capital gains by pension funds is notsubject to tax at either the entity or household level.


Figure 2.Realized Net Long-Term Gains and Corporate Equity ofHouseholds, 1954-1985 (In billions of dollars)

200

150

100

50

CorporateEquity

(Right Scale)

Long-termGains

(Left Scale)

2,000

1,500

1,000

500

1955 1960 1965 1970

Year

1975 1980 1985

SOURCES: Congressional Budget Office, computed from published data of the InternalRevenue Service and the Board of Governors of the Federal Reserve System.

The volatility of capital gains realizations has been greater for thetop income groups than for taxpayers as a whole (see Table 5 andFigure 3). During the 1970s, the ratio of realized gains to income forthe top 0.25 percent of taxpayers, ranked by AGI, fell to about one-third of its 1968 peak; the same ratio declined to only about two-thirdsof its 1968 peak for the bottom 99 percent of taxpayers.6/ After 1978,the growth in realizations as a share of GNP was much greater for thetop 1 percent (and top 0.25 percent) of returns than for the bottom 99percent.

Consequently, the share of realized gains in the top 0.25 percent,which ranged from 33 percent to 39 percent over the 1954 through

6. The division of the taxpaying population into percentiles of AGI is based on an interpolation ofpublished data on realizations by AGI group in the Internal Revenue Service publication, Statisticsoflncome.


TABLE 5. RATIO OF REALIZED NET LONG-TERM GAINS TO GROSSNATIONAL PRODUCT, BY INCOME GROUP, 1954-1985

Year

19541955195619571958195919601961196219631964196519661967196819691970197119721973197419751976197719781979198019811982198319841985

Top0.25 Percentof Returns(In percents)

0.70.90.80.60.70.90.81.10.80.81.01.10.91.21.51.20.70.81.00.70.50.40.50.50.50.90.80.91.21.41.51.7

Top1 Percentof Returns(In percents)

1.01.31.20.91.01.31.11.51.11.11.21.41.21.52.01.60.91.11.41.10.80.70.70.80.71.21.11.21.51.82.02.3

Bottom99 Percentof Returns(In percents)

0.91.11.11.01.11.31.21.41.21.31.41.61.61.82.01.81.21.41.61.61.31.31.51.51.51.61.41.41.21.61.61.8

SOURCES: Congressional Budget Office, computed from published data of the Internal Revenue Ser-vice and the Department of Commerce, Bureau of Economic Analysis.


1968 period, declined to a low of 21 percent in 1978, and then in-creased to between 40 percent and 42 percent in 1982 through 1985(see Table 6). The share of realized gains in the top 1 percent wasaround 50 percent for most of the 1950s and 1960s, dropping in the1970s to a low of 32 percent in 1978, and then increasing to around 55percent in 1982-1985.

RECENT HISTORY OF CAPITAL GAINS TAXATION

In recent years, changes in a number of provisions of the tax law haveaffected the taxation of long-term capital gains. These changes in-clude those made in:

Figure 3.Ratio of Realized Long-Term Gains to Gross National Product,by Income Group, 1954-1985

0.028

0.024

0.020

0.016

0.012

0.008

0.004

Ratio of Long-term Gains to GNP

Bottom99 Percent

1955 1960 1965 1970

Year

1975 1980 1985

SOURCES: Congressional Budget Office, computed from published data of the Internal RevenueService and the Department of Commerce, Bureau of Economic Analysis.

CHAPTER III CAPITAL GAINS TAXES AND REALIZATIONS, 1954-1985 31

TABLE 6. SHARE OF REALIZATIONS OF NET LONG-TERM GAINSBY THE HIGHEST-INCOME GROUPS, 1954-1985

Realized Net Long-Term Gains by Top1 Percent of Returns

Realized Net Long-Term Gains by Top

0.25 Percent of Returns

Year

19541955195619571958195919601961196219631964196519661967196819691970197119721973197419751976197719781979198019811982198319841985

Amount(In billionsof dollars)

3.65.25.03.94.46.45.78.26.56.67.99.99.4

12.617.915.79.2

12.516.414.411.410.412.715.115.831.131.336.748.162.173.992.0

Share(In percents)

51.254.152.147.247.049.548.852.547.845.746.647.343.446.250.048.143.344.445.440.438.034.032.434.132.243.644.346.955.252.954.355.6

Amount(In billionsof dollars)

2.63.73.62.73.04.64.26.14.95.06.37.77.39.7

13.212.0

6.79.1

11.710.07.57.08.8

10.510.522.522.026.537.347.457.669.9

Share(In percents)

36.938.437.532.732.535.535.638.835.834.636.737.233.735.437.036.831.532.132.228.124.922.722.423.621.431.531.133.842.840.442.842.2

SOURCE: Congressional Budget Office, computed from published data of the Internal RevenueService.


o The general structure of marginal rates and the level of per-sonal exemptions and standard deductions;

o The percentage of net long-term gains that may be deductedin computing taxable income;

o Provisions that have allowed taxpayers to use an alternativetax rate for capital gains that, for some taxpayers, was lowerthan the marginal rate multiplied by the percentage of gainsincluded in taxable income;

o Provisions that imposed minimum taxes on preference in-come, including the untaxed portion of long-term capitalgains, and that reduced the benefits of the maximum tax onearned income for taxpayers with preferentially taxed gains;and

o The holding period determining when a gain is eligible forlong-term treatment.

All of these changes have influenced the effective tax rate on reali-zations of long-term gains.

The fluctuations of gains between 1954 and 1985 have to somedegree mirrored the changes in tax rates applied to gains. Basically,capital gains taxation has gone through three distinct periods in the1954 through 1985 interval.?/ The first, which lasted until the TaxReform Act of 1969 and maintained capital gains provisions that hadbeen in effect since 1942, was marked by low and relatively stable taxrates on capital gains. The second period, between 1969 and 1978, wasmarked by high and rising effective tax rates on capital gains enactedin the 1969 act and the Tax Reform Act of 1976. The third period,between 1978 and 1985, was characterized by low and declining taxrates on capital gains enacted in the Revenue Act of 1978 and the Eco-nomic Recovery Tax Act of 1981. A new era, marked by higher rateson capital gains than in the recent past, and historically low rates onordinary income, is now beginning, following the 1986 act.

7. A full description of these changes is provided in Office of the Secretary of the Treasury, Office ofTax Analysis, Capital Gains Tax Reductions, pp. 29-39.


1954-1969

Between 1954 and 1969, 50 percent of long-term capital gains wasdeductible in computing taxable income. Absent other provisions, thisdeduction would have made the top rate on capital gains equal to 45.5percent (half of 91 percent) before the tax cuts enacted in the RevenueAct of 1964, and 35 percent (half of 70 percent) after the 1964 act wasfully phased in. In addition, however, long-term capital gains were al-lowed an alternative tax rate of 25 percent. In general, the maximummarginal tax rate on capital gains was 25 percent.8/

In 1968, the Congress enacted a 10 percent surtax to help financethe Vietnam war. The surtax was in effect between April 1,1968, andApril 1, 1970, and raised tax liability by 7.5 percent in 1968, 10 per-cent in 1969, and 2.5 percent in 1970. It raised the maximum mar-ginal tax rate on capital gains to 26.875 percent in 1968 and 27.5 per-cent in 1969.

1969-1976

The Tax Reform Act of 1969 included three major provisions affectinglong-term capital gains. First, it phased out the alternative tax over athree-year period for taxpayers with gains over $50,000. Removal ofthe limit on the alternative tax by itself raised the maximum tax rateon capital gains to 35 percent by 1972. Second, the act introduced anadd-on minimum tax on tax preference income above a specifiedexemption. The untaxed half of long-term capital gains was countedas a preference in computing the minimum tax. Finally, it introduceda new maximum tax of 50 percent on "earned income," compared with70 percent on "unearned income," but provided that the benefits of themaximum tax be reduced by 50 cents for every dollar of tax preferenceincome. This "poisoning" of the maximum tax also increased the taxrate on long-term capital gains for some taxpayers.

Some taxpayers could have confronted a marginal tax rate slightly higher than 25 percent. If ataxpayer was in a higher than 50 percent bracket with capital gains, but less than a 50 percentbracket with no capital gains, use of the alternative tax would mean forgoing some of theinframarginal benefits of lower statutory tax rates. See Lawrence B. Lindsey, Capital Gains: RatesRealizations and Revenues, Working Paper No. 1893 (National Bureau of Economic Research, Inc.,April 1986).


The effect of these provisions was to increase the marginal taxrate on long-term capital gains to 35 percent for taxpayers in the topincome bracket, to 36.5 percent for taxpayers in the top incomebracket subject to the minimum tax, and to 45.5 percent for taxpayersin the top income bracket affected by both the minimum tax and thepoisoning of the maximum tax.

The 1976 act further increased the taxation of capital gains byincreasing the rate of the minimum tax and reducing deductions fromthe minimum tax. These changes raised the maximum tax rate onlong-term gains to 39.875 percent for taxpayers in the 70 percentbracket and subject to the minimum tax and to a theoretical maxi-mum rate of 49.125 percent for taxpayers subject to the minimum taxand also affected by the poisoning of the maximum tax. (In practice,few taxpayers actually faced a 49 percent rate.) In addition, the 1976act increased the holding period required for a gain to be considered"long-term" from 6 months to 9 months in 1977 and 12 months in 1978and succeeding years.

1978-1985

The 1978 act substantially lowered the taxation of long-term capitalgains. It ended the "poisoning" of the maximum tax and removed thecapital gains preference from the base of the add-on minimum tax.9/The long-term capital gains deduction was increased from 50 percentto 60 percent. Finally, the alternative tax on capital gains was elimi-nated. The combined effect of these provisions was to lower the maxi-mum marginal tax rate on long-term gains to 28 percent.

The 1981 act reduced the top rate on ordinary income from 70 per-cent to 50 percent, effective January 1, 1982. This change by itselflowered the maximum tax rate on capital gains to 20 percent. Aspecial provision of the 1981 act made the 20 percent maximum rateon long-term capital gains effective for gains realized after June 20,1981. The 1981 act also generally lowered marginal tax rates acrossthe board over a three-year period, thereby reducing capital gains tax

9. To substitute for the add-on minimum tax on gains, the 1978 act introduced an alternativeminimum tax with a maximum tax rate of 25 percent. The maximum rate on the alternativeminimum tax was lowered to 20 percent in 1981.

CHAPTER HI CAPITAL GAINS TAXES AND REALKATIONS, 1954-1985 35

rates for taxpayers below the top income group. The Tax Reform Actof 1984 reduced the holding period to six months for gains realizedbetween December 22,1984, and January 1,1988.

The Tax Reform Act of 1986

The 1986 act eliminated the capital gains deduction, while at the sametime reducing marginal tax rates on ordinary income. For 1987, themaximum tax rate on capital gains was 28 percent. For the yearsafter 1987, the marginal tax rate on long-term capital gains is thesame as the rate on ordinary income. For taxpayers with the highestincomes, the maximum rate is 28 percent, the same as the rate in ef-fect after the 1978 tax reduction. For taxpayers just below the highestincomes, the effective marginal tax rate on both ordinary income andcapital gains is 33 percent because these taxpayers lose the benefits ofthe 15 percent bracket and personal exemptions at a rate of 5 cents perdollar of additional income.

Eliminating the capital gains deduction affects other features ofcapital gains taxation. The holding period for determining long-termgains becomes much less important because both net short-term andnet long-term gains are taxed at the same rate. This change removesthe incentive to delay realization of gains until they become long-termand also removes an incentive to claim short-term losses. Because thepreference for long-term capital gains is eliminated, long-term gainsare no longer affected by the alternative minimum tax. Finally, the1986 act significantly increased marginal rates on long-term capitalgains for low- and middle-income taxpayers, because these taxpayersreceive little or no reduction in statutory marginal tax rates on ordi-nary income. (The reductions in individual tax liability at the bottomof the income distribution resulted mainly from increases in personalexemptions and the standard deduction.)

Trends in Marginal Tax Rates

The maximum marginal tax rate on long-term gains was 25 percentbetween 1954 and 1967. It increased slightly in 1968 and 1969 withthe Vietnam War surtax, and increased in several steps between 1969and 1978 as a result of the 1969 act and the 1976 act (see Table 7). The


TABLE 7. MAXIMUM MARGINAL TAX RATE ON CAPITALGAINS, 1954-1988 (In percents)

Ordinary orOrdinary or Alternative Plus

Ordinary or Alternative Plus Minimum PlusYear Alternative Tax Minimum Tax Maximum Tax

1954195519561957195819591960196119621963196419651966196719681969197019711972197319741975197619771978 ^1979198019811982198319841985198619871988

25.025.025.025.025.025.025.025.025.025.025.025.025.025.026.927.530.232.535.035.035.035.035.035.033.828.028.023.720.020.020.020.020.028.028.0

25.025.025.025.025.025.025.025.025.025.025.025.025.025.026.927.532.334.336.536.536.536.539.939.939.028.028.023.720.020.020.020.020.028.028.0

25.025.025.025.025.025.025.025.025.025.025.025.025.025.026.927.532.338.845.545.545.545.549.149.148.328.028.023.720.020.020.020.020.028.028.0

SOURCE: Compiled by Congressional Budget Office. See also Office of the Secretary of the Treasury,Office of Tax Analysis, Report to Congress on the Capital Gains Tax Reductions of 1985(September 1978).

NOTES: In 1978, the maximum rate was reduced from 35 percent to 28 percent for gains realized afterOctober 31,1978.

In 1981, the maximum rate was reduced from 28 percent to 20 percent for gains realized afterJune 20,1981.

The 28 percent rate in 1988 applies to the highest-income group. For taxpayers with income inthe range in which the benefits of personal exemptions and the 15 percent rate are phased out,the marginal tax rate on capital gains is 33 percent.

CHAPTER in CAPITAL GAINS TAXES AND REALIZATIONS, 1954-1985 37

maximum rates then declined in 1978 and 1981, reaching a low pointof 20 percent for 1982 through 1986. The maximum rate was in-creased to 28 percent in 1987.

The average marginal rate on long-term gains for all returns dif-fers from the maximum rate because most taxpayers are not in the toprate bracket and because taxpayers with small amounts of gainsand/or low marginal tax rates were exempt from limits on the alterna-tive tax, the add-on minimum tax, and the poisoning of the maximumtax in years when those provisions were in effect. The average mar-ginal rate on capital gains can change even when the maximum rate isnot changing, for several reasons. First, legislated changes in thestructure of marginal tax rates on ordinary income alter the tax rateson capital gains in lower brackets by different amounts than the maxi-mum rate. Second, increases in nominal income move taxpayers be-low the very top income groups into higher marginal tax brackets,while leaving the maximum rate unchanged.!^/ Third, changes in thedistribution of asset holdings, as estimated by income flows, affectweights used to aggregate the separate tax rates of different incomegroups into an average rate. 117

The average marginal tax rate on long-term capital gains for alltaxpayers declined slightly between 1954 and 1965, then began risingto a peak of 22.7 percent in 1978 (Table 8 and Figure 4). It droppedsharply following the 1978 act and dropped again after the 1981 act toa low point of 13.9 percent in 1985.

10. The indexing of tax brackets in 1985 and 1986 kept taxpayers from being pushed into higherbrackets by changes in the price level, although there could still be "bracket creep" resulting fromreal economic growth. No indexing is in effect for 1987 and 1988, as the 1986 act changes in rates,exemptions, and bracket widths are phased in.

11. Average marginal tax rates on capital gains are computed as weighted averages of last-dollarmarginal tax rates on capital gains faced by taxpayers with the average amounts of capital gainsand taxable income in each income group. The income groups used are those reported by theInternal Revenue Service in published volumes of the Statistics of Income between 1954 and 1985.The weights used to aggregate among income groups are predicted long-term capital gains, withgains predicted by an equation that includes as explanatory variables dividends and adjusted grossincome less gains. The reason for using predicted instead of actual gains in the weighting is toremove any influence tax changes might have in altering the distribution of gains among incomegroups, thereby affecting the weights used in computing the average marginal rate. Usingpredicted instead of actual weights keeps the calculated average tax rate from being affected by theresponse of capital gains realizations to tax changes.


TABLE 8. ESTIMATED WEIGHTED AVERAGE MARGINAL TAXRATE ON CAPITAL GAINS, 1954-1985 (In percents)

Year

19541955195619571958195919601961196219631964196519661967196819691970197119721973197419751976197719781979198019811982198319841985

Projected 1988 Rates:Pre-1986 LawCurrent Law

AllReturns

17.317.718.017.217.317.116.717.116.816.916.216.116.216.718.618.819.519.920.119.519.520.121.922.222.718.118.616.814.814.414.013.9

14.425.4

Top1 Percent

23.623.924.323.823.824.023.924.024.024.123.723.824.024.226.227.127.928.930.030.229.930.233.834.435.125.926.123.420.019.719.419.5

19.728.7

Bottom99 Percent

9.19.49.89.79.89.59.810.010.110.59.99.59.610.111.311.711.911.611.511.211.212.112.813.213.711.111.911.610.69.99.79.6

10.423.0

SOURCE: Congressional Budget Office, computed from published data of the Internal RevenueService.

CHAPTER ffl CAPITAL GAINS TAXES AND REALIZATIONS, 1954-1985 39

The movement of the average rate on gains reflects slightly dif-ferent patterns of movement for the top 1 percent and bottom 99percent of returns. The tax rate on the top 1 percent of returns fluctu-ated in a narrow band between 23.6 percent and 24.3 percent between1954 and 1967. The rate then increased every year, with the exceptionof 1974, reaching a peak of 35.1 percent in 1978. It declined to 25.9percent in 1979, increased slightly in 1980, and then dropped to 20.0percent in 1982 and 19.4 percent in 1984.

For the bottom 99 percent, the rate increased gradually from 9.1percent in 1954 to 10.5 percent in 1963 because of the movement oftaxpayers into higher brackets induced by the growth in nominalincome. It then dropped to 9.5 percent in 1965 as a result of the 1964tax cut, but began rising again to 11.9 percent in 1970 as higher infla-

Figure4.Average Marginal Tax Rates on Long-Term Gains forSelected AGI Groups (Weighted by predicted gains)

Marginal Tax Rate on Long-term Gains

1955 1980 1985

SOURCE: Congressional Budget Office, computed from published data of the InternalRevenue Service.


tion pushed taxpayers into higher brackets. The rate dipped briefly inthe mid-1970s, but then increased further to a peak of 13.7 percent in1978, after which it declined with the 1978 capital gains cut and thephased-in 1981 marginal tax rate cut, reaching 9.6 percent in 1985,the lowest rate since 1966.

Under pre-1986 law, the average marginal rate on capital gainswould have risen slightly to 19.7 percent for the top 1 percent and 10.4percent for the bottom 99 percent in 1988. The 1986 act increasesthese rates in 1988 to 28.7 percent and 23 percent, respectively—an in-crease of 46 percent for the top 1 percent and 121 percent for thebottom 99 percent. The average marginal rate for the top 1 percent in1988 is slightly above the rate that group paid after the 1978 tax cutand below the rate it paid for most of the 1970s. On the other hand, forthe bottom 99 percent, the average marginal tax rate on capital gainsrealizations is substantially above any rate that group paid during theentire period.

MARGINAL TAX RATES AND REALIZATIONSOF GAINS: A FIRST GLANCE

Over the 1954-1985 period, realized gains have generally increasedmore than GNP when marginal tax rates on gains were declining, andincreased less than GNP when tax rates were rising and high. Thegains-to-GNP ratio increased gradually between 1954 and 1967, asthe marginal tax rate was declining slightly (see Figure 5). It rosesharply in 1968, a year the tax rate also increased. It then droppedsharply in 1970 and generally remained low until 1978, while taxrates were high and rising. After 1978, the gains-to-GNP ratio in-creased as the average marginal tax rate on gains declined. Thus,much of the divergence of realizations from GNP follows the fluctua-tions in tax rates. However, some of this fluctuation in realizationscan also be explained by movements in corporate equity values, andother nontax variables. In Chapter IV, the determinants of thesechanges are estimated statistically.


REVENUE FROM CAPITAL GAINS TAXES

Revenue from taxes on long-term capital gains depends both on thelevel of realized long-term gains and on the rate at which those gainsare taxed. As discussed above, realizations are influenced by the taxrate on gains, but also by other factors. The average tax rate is itselfaffected by the distribution of capital gains realizations: when theshare of realizations of upper-income groups rises, the average taxrate on gains rises even if the tax law is unchanged.

Figure 5.Marginal Tax Rates on Long-Term Gains and the Ratioof Long-Term Gains to Gross National Product (In percents)

5.0

4.5

4.0

3.5

3.0

2.5

2.0

1.5

Marginal Tax Rateon Long-term Gains

(Right Scale)

Ratio of Long-termGains toGNP(Left Scale)

25.0

- 22.5

20.0

17.5

15.0

1955 1960 1965 1970 1975

Year

1980 1985

SOURCES: Congressional Budget Office, computed from published data of the Internal Reve-nue Service and the Department of Commerce, Bureau of Economic Analysis.


Individual income tax assessments are based on taxable income,which is the sum of capital gains in adjusted gross income and othersources of taxable income, less deductions and exemptions. There isno separate tax schedule for capital gains. Individual tax liabilitiesattributable to long-term gains cart, however, be computed by sub-tracting from total individual liabilities an estimate of taxes thatwould be payable if long-term gains were zero.12/ Such a calculationshows that revenue attributable to taxes on long-term gains hasgrown over time from less than $1 billion in 1954 to almost $24 billionin 1985, the last year for which detailed individual tax return data areavailable (see Table 9). Over this interval, tax revenues from gainshave accounted for between 3.1 percent and 7.3 percent of total indi-vidual income tax liabilities.

The ratio of tax revenues from gains to total individual liabilitieshas fluctuated considerably over the 1954-1985 period. Revenuesfrom gains increased rapidly in the 1950s and 1960s, peaking at 6.9percent of individual liabilities in 1968. They declined to 3.6 percentof individual liabilities in 1970, rebounded to 5.8 percent in 1972, andfell again to 3.3 percent in 1974. Between 1976 and 1982, revenuesfrom gains varied between 4 percent and 5 percent of total individualliabilities. As a percentage of total liabilities, they increased sharplyalong with the stock market boom of the 1980s to 6.1 percent in 1983and a historical peak of 7.3 percent in 1985.

As a percentage of GNP, individual tax revenues from gains haveranged from 0.24 percent in 1957 to a peak of 0.59 percent in 1968 andagain in 1985. Revenue from gains taxes as a percentage of GNPincreased immediately following the 1978 capital gains tax cut, from0.38 percent in 1978 to 0.42 percent in 1979, but fell back to 0.40 per-cent in 1980 and 0.37 percent in 1981 and 1982. After 1982, however,this ratio increased sharply, rising to 0.49 percent in 1983, 0.53 per-cent in 1984, and 0.59 percent in 1985.

12. This method of computing revenues attributable to capital gains is somewhat arbitrary becausegains are "stacked last." In a progressive income tax, the assumption that gains are the "marginal"source of income means that the average tax ra|te on gains is equal to or greater than the averagetax rate on other income for every taxpayer. If tains were stacked first in the computation-that is,if taxes were computed on gains income alone, with other income set to zero-then the revenueattributable to gains would be smaller.

CHAPTER III CAPITAL GAINS TAXES AND REALIZATIONS, 1954-1985 43

TABLE 9. REVENUE ATTRIBUTABLE TO TAXES ONLONG-TERM CAPITAL GAINS, 1954-1985

Year

19541955195619571958195919601961196219631964196519661967196819691970197119721973197419751976197719781979198019811982198319841985

Revenuefrom Long-Term Gains(In billionsof dollars)

0.91.41.31.11.21.81.62.31.92.02.42.82.83.65.35.03.04.15.45.24.14.36.37.78.5

10.611.011.411.816.520.123.7

TotalIndividual

Tax Liabilities(In billionsof dollars)

26.729.632.734.434.338.639.542.244.948.247.249.556.162.976.686.683.885.293.4

107.9123.5124.4140.8158.5186.7213.3249.1282.3276.1271.7301.9325.6

Gains Revenuesas Share of:

IndividualLiabilities

(In percents)

3.54.64.03.13.64.64.15.54.24.25.05.74.95.86.95.83.64.85.84.83.33.54.54.84.54.94.44.14.36.16.77.3

GNP(In percents)

0.250.330.310.240.270.360.310.430.330.340.370.400.360.450.590.520.300.370.450.380.280.270.360.390.380.420.400.370.370.490.530.59

SOURCE: Congressional Budget Office, based on data on revenues from taxes on long-term and short-term gains supplied by U.S. Department of the Treasury, Office of Tax Analysis.


These data reveal no clear relationship between tax rates and taxrevenues; revenue from taxes on gains as a share of GNP increasedgreatly in the 1960s when tax rates were stable; declined but thenpartly recovered in the 1970s when capital gains tax rates were in-creasing; and remained flat for a time, but then increased sharply inthe late 1970s and early 1980s, when the tax rates were cut. In Chap-ter IV, the statistical estimates of the response of realizations to taxrates are used as an input to simulations of the revenue effects ofchanging the tax treatment of capital gains.

CHAPTER IV

STATISTICAL EVIDENCE ON CAPITAL

GAINS AND TAX REVENUES

Changes in the realizations of capital gains over the 1954 through1985 period can be explained by a simple equation in which realizedgains are represented as depending on the level and rate of change ofgross national product, the value of corporate equity held by individ-uals, and the marginal tax rate on capital gains. This chapter pre-sents statistical estimates of several variants of such an equation anduses the results of those estimates to simulate the revenue effects ofchanges in tax rates on capital gains.

There are many alternative ways of specifying equations for capi-tal gains realizations. The estimated behavioral response of realiza-tions to tax policy changes is sensitive to how the equation explainingrealizations is specified. A more detailed discussion of theoretical con-siderations in specifying these equations, along with estimates ofsome alternative specifications, are provided in Appendix A. Appen-dix A also contains a discussion of how data considerations limit theextent to which all possible hypotheses can be tested. Appendix Bdiscusses in more general terms some of the limitations and qualifica-tions that apply to all of the econometric work on realizations of capi-tal gains, including this study.

DETERMINANTS OF CAPITAL GAINS REALIZATIONS

Taxpayers may wish to sell assets and realize gains either to re-arrange their financial portfolios or to finance additional consumptionor investment in housing and consumer durables. The motives forrearranging their portfolios may be either a belief that the futureprospects of other assets are better than indicated by market prices, ora desire to reduce overall risk by selling assets that have recently


accrued large capital gains.U In either event, the total amount ofgains realized for portfolio purposes is likely to vary positively withthe available stock of accrued gains and negatively with the trans-actions cost of realizing gains.

Taxpayers may also realize gains to finance real investment orconsumption, as in purchases of houses and automobiles. Older tax-payers, in particular, may sell off part of the gains accumulated onsavings during their working years to finance consumption in retire-ment. For these reasons, the level of realized gains may be expected tovary positively with the total level of economic activity, as measuredby GNP.

Accrued Gains

The principal determinant of the level of realizations by individualtaxpayers as reported on tax returns is the stock of accrued gains onassets that are subject to the capital gains tax. Gains subject to taxupon realization include those accrued in the current year and pre-vious years on corporate equities, bonds, ownership of shares in pro-prietorships and partnerships, land, and on gold, art, and other collect-ibles. They do not include gains on assets held indirectly on behalf ofindividuals by pension funds and life insurance companies, becauseincome of these entities is not subject to tax. They also largely excludeaccrued gains on owner-occupied housing because most realized gainson owner-occupied housing are not taxed.2/ Accrued gains cannot beobserved directly because data are not available on the basis of assetsin the economy (see Box 1). In the equations reported in this chapter,a measure of household wealth is used as a proxy for accrued gains.The measure of wealth is the end-of-year total value of corporateequities held by individuals, published by the Board of Governors of

1. For evidence that average rates of returns on realized stock market gains are much higher than theaverage growth of the market as a whole, see Office of the Secretary of the Treasury, Report toCongress on the Capital Gains Tax Reductions of 1978 (September 1985), pp. 69-75.

2. Capital gains on owner-occupied housing are largely untaxed for three reasons: gains can normallybe rolled over if a more expensive home is purchased; since 1978, $125,000 of gains are excused onone sale for a taxpayer aged 55 or over; and there is a step-up in basis on homes (as well as otherassets) transferred at death.

CHAPTER IV STATISTICAL EVIDENCE ON GAINS AND REVENUES 47

BOX1ESTIMATING ACCRUED GAINS

When someone sells an asset, the capital gain (or loss) is equal to the dif-ference between the total sales value of the asset and its cost or "basis." Thebasis of the asset is its initial purchase price plus selling costs plus, in the caseof real property, the cost of improvements.

Published data are available on the total market value of certain types ofassets held by households, such as corporate equities. There are no compara-ble data on what the owners paid for their assets. Therefore, the amount of ac-crued gains, which is the difference between asset value and basis, cannot beobserved directly.

The estimation of accrued gains is difficult, for several reasons. A num-ber of influences can change the ratio of basis to value for assets in the econ-omy, thereby changing the ratio of wealth to accrued gains. For example, if in1960 taxpayer A sold $1,000 worth of shares in company I to taxpayer B andused the proceeds to purchase $1,000 worth of shares in company II from tax-payer B, this transaction itself would not change the total wealth held by thetwo taxpayers. It would, however, increase their basis in 1961 if the assets hadoriginally been purchased for less than $1,000, and reduce their basis in 1961if the assets had been purchased for more than $1,000. Put another way, therealization of a capital gain, followed by a subsequent repurchase, increasesthe ratio of basis to wealth and reduces the ratio of accrued gains to wealth.The realization of a capital loss has the opposite effect.

Several factors could contribute to changes in the ratio of accrued gains towealth. Appreciation in the value of existing assets raises the ratio of accruedgains to wealth. When taxpayers realize more capital gains, the ratio of ac-crued gains to wealth declines. Similarly, when a taxpayer dies and passes onan asset with accrued gains to a beneficiary, the basis of the asset is "steppedup" from its original purchase price to its market value at the time it is in-herited. This also reduces the ratio of accrued gains to wealth.

Finally, under the federal income tax, it has often been possible to depre-ciate assets for tax purposes more rapidly than their actual decline in marketvalue. Depreciation deductions reduce the basis of assets in the same way thatthe realization of gains increases the basis. Thus, when tax depreciation is ac-celerated, the ratio of basis to wealth declines and the ratio of accrued gains towealth increases.

82-667 0 -


the Federal Reserve System.3/ Corporate equities are a significantfraction of assets that give rise to capital gains, and also generally canbe sold at lower transactions cost than other capital gains assets. Nomeasure of wealth, however, includes the effect that changes in thebasis may have on realizations of gains (see Box 1).

Equations were also estimated in which accrued gains wererepresented by a linear combination of all the variables that arebelieved to influence either wealth or basis, and in which a. compre-hensive accrued gains variable was created by making assumptionsabout the missing data. These equations are reported in Appendix A.

Total Economic Activity

GNP was used in the equations as a measure of total economicactivity. Alternative measures, such as national income or personalincome, would yield much the same results in the equations.

The use of GNP as an explanatory variable captures two possibleeffects on capital gains realizations. First, GNP is the most compre-hensive measure of total spending, and in part captures the motive tosell capital gains assets either for consumption or for investment inhousing and consumer durables. Second, as a measure of total eco-nomic activity, GNP may capture some of the influences on totalwealth not included in the stock market variable. When combinedwith stock values, GNP performs better in explaining capital gainsthan a measure of household wealth in noncorporate equity .47

3. See Board of Governors of the Federal Reserve System, Balance Sheets for the U.S. Economy (May1987).

4. The estimated value of noncorporate equity held by households, including farms and rental realestate, is also published by the Board of Governors of the Federal Reserve System. Fornoncorporate equity, however, the Federal Reserve Board data are based on an estimate of thereplacement value of capital, not on actual market prices. For corporations, where both figures areavailable, the ratio of market value to the estimated replacement cost of capital has fluctuatedconsiderably over time. Therefore, the estimated value of noncorporate equity is not a very goodproxy for accrued gains.


Marginal Tax Rates

The main concern of this study is the extent to which permanentchanges in marginal tax rates on long-term gains affect realizations ofcapital gains, for any given level of capital gains accruals. In a grow-ing economy, the long-run ratio of realizations to accruals can increaseeither if the average holding period of assets is shortened (that is, ifassets are sold more frequently) or if taxpayers realize a larger frac-tion of the gains accrued during their total lifetimes instead of passingthem on tax-free to their beneficiaries.

Higher marginal tax rates on realized gains can both lengthenaverage holding periods and reduce the percentage of accrued gainsrealized during a taxpayer's lifetime. In particular, for taxpayers re-arranging their portfolios, the overall benefits in terms of higherfuture expected returns or lower risk must be high enough to offsettransactions costs, including capital gains taxes. The possibility thatmore asset sales will be desirable rises if the capital gains tax islowered, leading to a more rapid turnover of portfolios.

It can be shown mathematically that even a substantial reductionin turnover (or a substantial increase in the average holding period)will have a relatively small negative impact on the long-run ratio ofrealizations to accruals, although it has a considerable first-year effectand also lowers the present value of realizations. This smallerlong-run impact occurs because an immediate delay of realizationsavoids an increase in the tax basis of assets, thereby raising potentialfuture realizations. In a growing economy, the net effect is that alengthening of the holding period permanently lowers the ratio ofrealizations to accruals by a modest amount.5/ In the absence ofgrowth, less frequent turnover would have no effect on the ratio ofrealizations to accruals in the long run because smaller realizations inany one year would be exactly offset by larger realizations in thefuture.

A much larger permanent tax effect can be attributed to the step-up in basis at death. A high tax rate on gains realized during a tax-

5. Martin J. Bailey, "Capital Gains and Income Taxation," in Arnold C. Harberger and M.J. Bailey,eds., The Taxation of Income from Capital (Washington, D.C.: Brookings Institution, 1969), pp. 11-49.


payer's lifetime provides a large incentive to avoid consumption out ofassets with accumulated gains and to hold onto those assets to leave asbequests. Because a large fraction of capital gains is realized by tax-payers in the very highest income groups—those who can be expectedto leave the largest bequests—a big differential between the rate ongains realized during a taxpayer's lifetime and the rate on gainspassed on at death (currently zero) can have a significant, negativeeffect on the permanent level of realizations.6/

The equations in this chapter use the current marginal tax rate onlong-term gains as the only tax variable explaining realizations. Ifthe short-run and long-run effects of tax rate changes on realizationswere in fact the same, the coefficient on tax rates estimated from theseequations could be properly interpreted as a measure of both the long-run and short-run effects of changes in tax rates. If the long-run andshort-run effects differ, however, use of the current tax rate alonegives biased estimates of both. If the long-run effect is smaller thanthe short-run effect, the coefficient on the current tax rate understatesthe absolute size of the short-run effect but overstates the size of thelong-run effect. On the other hand, if the long-run effect is larger thanthe short-run effect because, for example, taxpayers are slow torespond to changes in incentives, then the coefficient on the currenttax rate overstates the absolute size of the short-run effect but under-states the size of the long-run effect.7/

The data do not permit estimation of a complex set of laggedrelationships between tax rates and realized gains. Experimentationwith a one-period lag on the tax rate variable, shown in Appendix A,produced inconclusive results. In one specification, the long-run effectwas estimated to be larger than the short-run effect, and in anotherspecification the reverse was found. In neither case was the coefficienton the lagged tax rate term statistically significant.

6. The differential treatment of gains realized during a taxpayer's lifetime and gains left as bequestscan be reduced either by lowering the tax rate on capital gains realizations or by eliminating thestep-up in basis at death. Elimination of step-up can be accomplished either by requiringbeneficiaries to inherit the tax basis of the assets of decedents (carryover basis) or by treating deathas a "realization event" and taxing gains not realized during a taxpayer's lifetime. A provision forcarryover basis, with a very long transition, was enacted in the 1976 act, but repealed in 1980.

7. For a fuller discussion of the bias in eliminating lag terms, see Appendix A.


STATISTICAL ESTIMATES

This section presents the major statistical results of this study. Itdescribes the equations that estimate how much capital gains reali-zations are affected by changes in marginal tax rates on capital gains.The section that follows presents results of simulations of revenuefrom capital gains taxation that use the estimated responses of howrealizations of capital gains respond to changes in marginal tax rates.The following section also discusses how statistical uncertainty in theestimates of the response of realizations causes uncertainty about thesize and direction of revenue effects. An evaluation of the results innontechnical terms is presented in the final section of this chapter.

Equations for the Entire Population of Taxpayers

Table 10 presents the results of four equations explaining the growthin long-term capital gains realizations between 1954 and 1985 for theentire taxpaying population. In all four equations, the dependentvariable is the logarithm of either nominal or real long-term capitalgains reported on all individual income tax returns. The independentvariables are the price level, GNP, the change in GNP, the value ofcorporate equities held by individuals, and the marginal tax rate oncapital gains. GNP, the change in GNP, and the value of corporateequities are measured in constant dollars. Except for the marginal taxrate, the independent variables are in logarithmic form. The log-arithmic form is chosen to represent the approximately exponentialgrowth rate in all the variables. The semi-log relationship betweenrealizations and tax rates means that each percentage-point change inthe tax rate results in the same percentage change in realized gains.Thus, for example, an increase in the tax rate from 0 to 1 percentwould cause the same percentage reduction in realizations as an in-crease from 29 percent to 30 percent.8/

If the tax rate were also entered in log form, then each percent increase in the tax rate would resultin the same percent reduction in capital gains. Under that functional form, an increase in themarginal tax rate from 1 percent to 1.5 percent would result in the same percent decline inrealizations as an increase from 30 percent to 45 percent.

(Continued)


TABLE 10. REGRESSION EQUATIONS EXPLAINING CAPITAL GAINSREALIZATIONS FOR ALL TAXPAYERS, 1954-1985

(1) log(LTG) = -7.874 + 1.103*log(PRICE) + 0.537*log(RCOEQ)(3.63)a (4.96)a (4.11)a

+ 1.128*log(RGNP) - 0.032*MTR(3.09)a (2.33)b

R2 = .985 S.E.E. = .113 D.W. = 1.559 T* =31.3%

(2) log(LTG) = -6.822 + 1.173*log(PRICE) + 0.501*log(RCOEQ)(3.38)a (5.75)a (4.19)a

+ 1.022*log(RGNP) + 2.067*dlog(RGNP)-0.031*MTR(3.05)a (2.57)b (2.47)b

R2 = 989 S.E.E. = .102 D.W. = 1.404 T* =32.3%

(3) log(RLTG) = -8.810 + 0.489*log(RCOEQ) + 1.293*log(RGNP)(11.5)a (6.09)a (18.5)a

- 0.037*MTR(3.95)a

R2 = .940 S.E.E. = .111 D.W. = 1.550 T* =27.0%

(4) log(RLTG) = -8.420 + 0.424*log(RCOEQ) + 1.301*log(RGNP)(11.7)a (5.42)a (20.2)a

+ 1.975*dlog(RGNP)-0.039*MTR(2.494)b (4.517)a

R2 = .951 S.E.E. = .102 D.W. = 1.417 T* =25.6%


NOTES: t-statistics are in parentheses.a = significantly different from zero at 1% level,b = significantly different from zero at 5% level.T* = revenue-maximizing tax rate.

Variable Definitions:

log(LTG) = logarithm of net long-term gains, in excess of net short-term losses;log(PRICE) = logarithm of the GNP deflator;log(RCOEQ) = logarithm of the value of corporate equities held by households (in constant

dollars);log(RGNP) = logarithm of GNP (in constant dollars);dlog(RGNP) = change in the logarithm of GNP (in constant dollars);MTR = weighted average marginal tax rate on capital gains; andlog(RLTG) = logarithm of net long-term gains, in excess of net short-term losses (in constant

dollars).


In Equation 1, capital gains in current dollars are represented asdepending on stock market values and GNP (in constant dollars), theprice level, and the marginal tax rate on capital gains. Realizationsincrease with increases in the overall price level, GNP, and the valueof corporate equities held by individuals, and decrease with increasesin the marginal tax rate on gains. Movements in these four variablesexplain 98.5 percent of the year-to-year changes in realized gains overthe 1954-1985 period (see the R2 statistic in Table 10). There remains,however, a considerable unexplained component of capital gains; thestandard error of estimate (S.E.E.) implies that about 30 percent of thetime gains will be at least 11 percent above or below the amount pre-dicted by the equation. The coefficient on the tax rate term MTR im-plies that a one-percentage-point increase in the average marginal taxrate on capital gains will reduce realized long-term gains by 3.2percent. For a flat-rate tax system, it implies that the revenue-maximizing tax rate T* on capital gains would be slightly over 31 per-cent (1/.032, as shown in Table 10).

Examination of the residuals of Equation 1 shows that it under-predicts gains in years of strong economic growth and overpredictsgains in recession years. In recent years, the equation overpredictsgains in 1980 through 1982 and underpredicts gains in 1983 through1985 (see Table 11). This result suggests inclusion of a cyclicalvariable in the equation. Equation 2 is similar to 1 except that thechange in the logarithm of real GNP (the percentage growth in realGNP) is added as an explanatory variable. Adding the growth rate ofGNP to the equation reduces the standard error, increases theexplanatory power of the equation, and produces a much better fit ofthe equation to data from recent years. The tax rate coefficient re-

8. Continued

The semi-log form is chosen instead of the log form because the same percentage increases in thetax rate will probably have a much bigger impact on gains if the tax rate is large to begin with. Thesemi-log form implies that the elasticity of realizations with respect to the tax rate increases inabsolute value as the tax rate is increased. This means that, starting from a zero tax rate, increasesin marginal tax rates at first raise revenues but then could ultimately lower revenues if theabsolute value of the elasticity grew to become greater than one. For a flat-rate tax on gains, therevenue-maximizing tax rate is -I/A, where A is the percentage reduction in gains per unitpercentage-point increase in the marginal tax rate.

An alternative functional form in which the tax rate measure used is the logarithm of the gain perdollar of pretax gain (1 minus the marginal tax rate) also has the property of yielding an elasticitythat increases in absolute value as the tax rate increases. Equations using this form, a quadraticform, and a logarithmic form are reported in Appendix A. The results are qualitatively similar tothose reported in Table 10.


mains virtually unchanged (-0.031 instead of-0.032, which implies arevenue-maximizing rate of 32 percent instead of 31 percent).

Equations 3 and 4 use the logarithm of gains in constant dollarsinstead of gains in current dollars as the dependent variable andeliminate the price level term as an independent variable. Thischange is equivalent to constraining the coefficient on the price termin Equations 1 and 2 to be equal to 1.0; that is, a 1 percent increase inthe price level raises gains, measured in current dollars, by 1 percentand has no effect on gains measured in constant dollars. Because theestimated coefficient on the price term in Equations 1 and 2 is veryclose to 1.0, one cannot reject the hypothesis that gains in constantdollars are unaffected by the price level.

The estimates of the responsiveness of realized gains to taxes aresomewhat higher in Equations 3 and 4 than in Equations 1 and 2. Forexample, Equation 4 implies that a one-percentage-point increase inthe marginal tax rate on capital gains will lower realizations oflong-term gains by 3.9 percent. This estimate is consistent with arevenue-maximizing flat rate of slightly under 26 percent. The othercoefficients of Equations 3 and 4 are similar to those of Equations 1and 2; again, including the change in the logarithm of GNP in the

TABLE 11. ACTUAL MINUS FITTED GAINS IN RECENT YEARSIN EQUATIONS FOR ALL TAXPAYERS (In percents)

Equation 1980 1981 1982 1983 1984 1985

(1)(2)

(3)

(4)

-6.2

-2.0

-6.0

-0.3

-9.0

-7.6

-7.6

-7.0

-11.0

-1.6

-10.0

-0.7

+ 4.5

+ 1.4

+ 5.0

+ 3.0

+ 6.8

-1.1

+ 7.0

-0.1

+ 8.8

+ 7.3

+ 8.2

+ 8.9



equation improves its predictive power and does a better job of fittingthe data for the early 1980s. The statistical evidence does not favoreither the higher tax responsiveness of Equations 3 and 4 or the lowertax responsiveness of Equations 1 and 2.97

Separate Equations for the Top Income Group and All Other Groups

Tables 12 and 13 report results for separate estimates of Equations 1through 4 for the top 1 percent and bottom 99 percent of tax returns,ranked by AGI. For the top 1 percent, the estimated reduction inrealizations for a one-percentage-point increase in marginal tax ratesranges between 2.9 percent and 3.2 percent, about the same as theestimated response for the entire sample in Equations 1 and 2. Thecoefficient on marginal tax rates is statistically significant at the 1percent level in all four equations. The implied revenue-maximizingflat-rate tax of between 31 percent and 34 percent is within the samplerange of tax rates for the top 1 percent.

For the bottom 99 percent, the estimates are much less precise. Inall four equations, the tax rate effect is negative, but it is not signifi-cantly different from either zero or the estimated effect for the top 1percent.10/

9. The tax response is slightly higher in Equations 3 and 4 because the price level is estimated to havea positive, though small, effect on constant-dollar gains (the coefficient of the price level inEquations 1 and 2 is slightly above 1.0). The tax rate is negatively correlated with the price leveland therefore picks up some of the price effect in Equations 3 and 4 when the price level is omitted.

The higher responsiveness found in Equations 3 and 4 would be preferred if random fluctuationsaccounted for the price coefficients being above 1.0. (In this case, the price level truly has no effecton gains (in constant dollars) and its absence from Equations 3 and 4 represents the correctspecification). On the other hand, the lower responsiveness of Equations 1 and 2 would be preferredif inflation actually has a small positive effect on gains. (In this case, the tax rate coefficients inEquations 3 and 4 are biased away from zero by the omission of the price level.) Statistical tests donot show one possibility to be much more likely than the other.

The estimated Durbin-Watson (D.W.) statistics in Table 10 fail either to confirm or to reject thehypothesis of no autocorrelation of residuals. Using the Cochrane-Orcott procedure for correctionfor autocorrelation does not significantly alter the estimated coefficients of the equations.Equations with an autoregressive term are displayed in Appendix A.

10. One reason for the failure to identify a significant tax effect for the bottom 99 percent may be thattheir marginal tax rates have varied only slightly during the sample period, while the variance inrates for the top 1 percent has been much larger. The fact that absolute changes in marginal taxrates on gains have been much larger for the top 1 percent than for the bottom 99 percent alsoexplains why the share of gains realized by the top 1 percent has declined when average marginaltax rates on gains have fallen (see Table 2).


TABLE 12. REGRESSION EQUATIONS EXPLAINING CAPITAL GAINSREALIZATIONS FOR THE TOP 1 PERCENT OF RETURNS,1954-1985

(1.01) log(LTGOl) = -10.900 + 0.901*log(PRICE) + 0.848*log(RCE01)(4.93)a (3.13)a (7.30)a

+ 1.839*log(RY01)-0.032*MTR01(4.08)a (5.81)a

R2 = .984 S.E.E. = .118 D.W. - 1.796 T* = 31.3%

(2 01) log(LTGOl) = -7.620 + 1.272*log(PRICE) + 0.899*log(RCE01)(3.27)a (4.34)a (8.47)a

+ 1.175*log(RY01) + 1.423*dlog(RY01)-0.029*MTR01(2.48)b (2.71)b (5.78)a

R2 = .987 S.E.E. = .106 D.W. = 2.243 T* = 34.5%

(3.01) log(RLTGOl) = -10.214 + 0.872*log(RCE01) + 1.688*log(RY01)(11.2)a (9.73)a (19.8)a

-0.032*MTR01(6.08)a

R2 = .943 S.E.E. = .116 D.W. = 1.81 T* = 31.3%

(4.01) log(RLTGOl) = -9.623 + 0.838*log(RCE01) + 1.607*log(RY01)(ll.l)a (10.1)a (19.2)a

+ 1.196*dlog(RY01)-0.031*MTR01(2.58)b (6.54)"

R2 = .954 S.E.E. = .106 D.W. = 2.169 T* = 32.3%


NOTES: t-statistics are in parentheses.a = significantly differentfrom zero at 1% level,b = significantly different from zero at 5% level.T* = revenue-maximizing tax rate.


log(LTGO 1) = logarithm of net long-term gains, in excess of net short-term losses, received by top1 percent of returns ranked by AGI;

log(RCE01) = logarithm of corporate equities (in constant dollars) held by individuals multipliedby share of dividends received by top 1 percent of returns;

log(RYOl) = logarithm of GNP (in constant dollars) multiplied by share of AGI other thancapital gains received by top 1 percent;

dlog(RYOl) = change from the preceding year in log(RYOl);MTR01 = average marginal tax rate on long-term capital gains for top 1 percent of returns;PRICE = GNP deflator; andlog(RLTG01) = logarithm of net long-term gains, in excess of net short-term losses, received by top

1 percent of returns (in constant dollars).


TABLE 13. REGRESSION EQUATIONS EXPLAINING CAPITAL GAINSREALIZATIONS FOR THE BOTTOM 99 PERCENT OFRETURNS, 1954-1985

(1.99) log(LTG99) = -11.709 + 0.725*log(PRICE) +0.260*log(RCE99)(6.19)a (4.72)a (2.86)a

+ 1.774*log(RY99)-0.032*MTR99(5.72)a (1.47)

R2 = .990 S.E.E. = .094 D.W. = 1.483 T* = 31.3%

(2.99) log(LTG99) = -10.825 + 0.796*log(PRICE) + 0.229*log(RCE99)(6.35)a (5.76)» (2.81)a

+ 1.677*log(RY99) + 1.868*dlog(RY99) -0.028*MTR99(6.06)a (2.90)a (1.40)

R2 = .992 S.E.E. = .084 D.W. = 1.461 T* = 35.7%

(3.99) log(RLTG99) = -8.435 + 0.376*log(RCE99) + 1.239*log(RY99)(16.8)«» (5.67)a (14.4)a

-0.008*MTR99(0.444)

R2 = .954 S.E.E. = .098 D.W. = 1.409 T* = 125.0%

(4.99) log(RLTG99) = - 8.378 + 0.310*log(RCE99) + 1.283*log(RY99)(19.1)a (5.04)a (16.8)a

+ 2.036*dlog(RY99) - 0.009*MTR99(3.14)a (0.59)

R2 = .967 S.E.E. = .085 D.W. = 1.342 T* = 111.1%


NOTES: t-statistics are in parentheses.a = significantly different from zero at 1% level,b = significantly different from zero at 5% level.T* = revenue-maximizing tax rate.


log(LTG99) = logarithm of net long-term gains, in excess of net short-term losses, received bybottom 99 percent of returns ranked by AGI;

log(RCE99) = logarithm of corporate equities (in constant dollars) held by individuals multipliedby share of dividends received by bottom 99 percent of returns;

log(RY99) = logarithm of GNP (in constant dollars) multiplied by share of AGI other thancapital gains received by bottom 99 percent;

dlog(RY99) = change from precedingyear in log(RY99).MTR99 = average marginal tax rate on long-term capital gains for bottom 99 percent of

returns;PRICE = GNP deflator; andlog(RLTG99) = logarithm of net long-term gains, in excess of net short-term losses, received by

bottom 99 percent of returns (in constant dollars).


The point estimates of the tax coefficient in Equations 1 and 2 are-0.032 and -0.028, almost the same as the point estimates for the top 1percent in the same equations. The point estimates fall almost to zeroin Equations 3 and 4, however. These levels are well below the taxcoefficients for the top 1 percent in those equations. The large drop inthe tax response in Equations 3 and 4 arises in part because the coeffi-cients of the price level in Equations 1 and 2 are significantly belowzero.:!!/ Those coefficients mean that inflation is estimated to lowergains (in constant dollars) significantly for the bottom 99 percent oftaxpayers. Thus, Equations 3 and 4, which assume that inflation hasno effect on constant-dollar gains, appear to be misspecified and theirestimates of tax effects biased. Equations 1 and 2, with their highertax responsiveness, are the preferred estimates for the bottom 99 per-cent of returns.

Even though the tax variable is less important for low-incomegroups, the equations fit better for the bottom 99 percent than for thetop 1 percent. In particular, realizations for the bottom group are lessvolatile and their movements track more closely the movements in theeconomy than do changes in realizations for the top group. In addi-tion, realizations in the bottom 99 percent of returns are much lessaffected by movements in stock market values than are realizationsfor the top 1 percent.

The data used for estimating regressions for the separate groupsare less reliable than those used for estimating the entire sample,because the data for separate groups are not exact. The division ofgains between the two groups must be interpolated because publisheddata classify income groups by nominal dollars instead of percentiles.The division of wealth holdings among income groups is not knownand must be imputed from published data on income flows. In addi-tion, a bias is introduced when splitting the data because the measureused to classify the top 1 percent of returns, AGI, itself depends on therealizations of gains. As realizations increase, taxpayers will be

11. A second reason the tax coefficient changes more for the bottom 99 percent is that their tax rate ismore strongly related to the price level than is the tax rate for the top 1 percent (or that for alltaxpayers). Tax rates for the top 1 percent are so weakly related to the price level that the taxcoefficients are largely unaffected even though the price coefficient is significantly above 1.0 inEquation 2. For all three groups the tax rate is negatively related to the price level, holding fixedequity and GNP (in constant dollars). For the top 1 percent, the relationship is smaller and notstatistically significant.


moved into the top 1 percent of returns because of capital gains so thatthe composition of the top 1 percent will change. The result is thatcapital gains in the top group may be overstated in years when totalgains are high, and understated in years when total gains are low,compared with the increase in gains in the top group that would be ob-served if the composition of the top group were not affected by inducedrealizations. A correction for this bias suggests it has only a slighteffect on the estimates. 12/

RELIABILITY OF THE ESTIMATES AND THEIRIMPLICATIONS FOR REVENUE EFFECTS

While the point estimates shown in Table 10 are all fairly close to-gether, it is possible to derive varying estimates by changing the spe-cification of the realizations equation. A wider range of alternativespecifications is displayed in Appendix A.

The estimates in Table 10 are not precise in themselves; the stan-dard error of the tax coefficient is big enough so that a wide range ofpossible responses cannot be rejected. For example, a 95 percent confi-dence interval (which includes values within roughly two standarderrors of the estimate) for the tax coefficient in Equation 1 includesvalues as small as -0.005 and as large as -0.059 (see Table 14). Thelower-bound response implies that any feasible tax rate increasewould raise revenue; the upper-bound response, on the other hand, im-plies that revenue is maximized at about a 17 percent tax rate. A 95percent confidence interval around the coefficients of Equations 3 and4 shows a lesser, but still considerable, range, with the implied reve-nue-maximizing rates ranging from 18 percent to around 50 percent.

The entire relationship between tax rates and revenue, not onlythe revenue-maximizing point, is highly sensitive to the estimate ofthe realizations response. Table 15 and Figure 6 show the relation-ship between revenue from capital gains taxes and the tax rate on

12. The correction in the data is based on more complete information available for 7 of the 31 years inthe sample, and imputed for the other years. Substitution of the corrected data in Equation 1 inTables 12 and 13 reduces the estimate of the tax response from -0.0323 to -0.0316 for the top 1percent and increases the estimated response from -0.0329 to -0.0348 for the bottom 99 percent.The method used to correct the data is described in Appendix A.


capital gains (again assuming a flat-rate tax) for the midpoint andextreme values of the 95 percent confidence interval around the reali-zations response estimated in Equation 1. The table and figure arecalibrated to set revenue equal to 100.00 for a tax rate of 20 percent(the maximum rate under pre-1986 law); the other numbers in thetable show revenue relative to revenue at a 20 percent rate.

For the most likely estimate of the realizations response inEquation 1 (the midpoint of the confidence interval), the revenue-maximizing rate is 31.3 percent, with revenue at that rate 9 percenthigher than revenue at a 20 percent rate. The behavioral response

TABLE 14. CONFIDENCE INTERVALS FOR TAX EFFECT ONREALIZATIONS AND REVENUE-MAXIMIZING RATEFOR ENTIRE POPULATION OF TAXPAYERS

95 PercentConfidence Interval

Equation Point Estimate on Realizations Response

Percentage Change in Realizations per Unit Changein Marginal Tax Rates

(1) -0.032 -0.005 to-0.059

(2) -0.031 -0.006 to-0.056

(3) -0.037 -0.019 to-0.055

(4) -0.039 -0.022 to-0.056

Revenue-Maximizing Flat Rate(In percents)

(1)(2)

(3)

(4)

31.2

32.3

27.2

25.8

16.9 to 197. 9

18.0 to 158.6

18.2 to 53. 8

18.0 to 45. 6

SOURCE: Congressional Budget Office


offsets much of the "static" gain from higher rates; with no behavioralresponse, revenue at 31.3 percent would be about 57 percent higherthan revenue at 20 percent. The revenue changes are very small forchanges in rates near the revenue-maximizing rate. For example, re-ducing the rate to 25 percent and increasing it to 40 percent both lowerrevenue by less than 5 percent.

For the upper-bound realizations response in the confidence inter-val, revenue peaks at a 16.9 percent rate, and revenue at a 15 percentrate is higher than at a 20 percent rate. Revenue falls off very sharplyas rates decline to 10 percent and below. Thus, although the point

TABLE 15. IMPLIED SHAPE OF RELATIONSHIP BETWEEN TAXRATES AND REVENUE FROM CAPITAL GAINS TAXES

20 100.0

Point on Confidence Intervalof Realizations Response

MarginalTax Rateon Gains

05101516.9

Mid-point(-.032)

040.468.9-88.093.3

HighResponse

(-.059)

060.690.2

100.7101.5

LowResponse

(-.005)

026.952.676.985.8

ZeroResponse

025.050.075.084.5

100.0 100.0 100.0

253031.335404550

106.5108.9109.0108.3105.5101.195.7

93.183.180.372.261.551.542.6

121.9142.7147.9162.4181.0198.6215.2

125.0150.0156.6175.0200.0225.0250.0


NOTE: This example assumes a flat-rate tax, using the realizations response estimated in Equation 1,showing midpoint and extreme values. Revenues are indexed at 100 for a tax rate of 20 percent.


estimate implies a considerable revenue loss from lowering the capitalgains tax rate to 15 percent, the possibility of a revenue gain cannot beentirely rejected. On the other hand, for the lower-bound response,revenue rises almost as much with higher tax rates as it would if therewere no behavioral response. For example, a doubling of the rate from

Figure 6.Relationship of Revenue from Capital Gains Taxes toMarginal Tax Rates on Gains

Revenue Index = 100 at 20 percent tax rate220

0 5 1 0 1 5 2 0 2 5 3 0 3 5 4 0 4 5 5 0

Marginal Tax Rate on Gains


NOTE: The curves show how revenues change with marginal tax rates under three responses ofrealizations to marginal tax rates: the response estimated in Equation 1 of Table 10;and the highest and lowest responses within a 95 percent confidence interval aroundthat estimated response.


20 percent to 40 percent raises revenue by 81 percent, compared with a100 percent increase with no behavioral response.

Table 16 compares annual revenue from capital gains taxes underpre-1986 law at 1988 income levels with two alternatives: currentlaw, and a proposal to amend current law by setting a maximum rateof 15 percent on long-term capital gains and removing long-term gainsfrom the alternative minimum tax on tax preference income. Revenuefrom capital gains taxes is calculated by subtracting from total indi-vidual income tax liabilities an estimate of total taxes that would bepayable if net long-term capital gains were zero. Revenue under pre-1986 law is estimated at the level of gains realizations that wouldhave been projected under previous law. (This estimate differs froman estimate of actual 1988 realizations because the 1986 tax law in-duced people to accelerate their gains from 1987, 1988, and subse-quent years to 1986 to avoid paying tax at the increased rates.) Reve-nue under alternative laws is estimated at simulated levels of realiza-tions; the changes in realizations are computed by multiplying thechanges in marginal tax rates on long-term gains (at the initial levelof realizations) in the two alternative laws by the realizations re-sponse coefficients estimated in Equations 1 through 4. (These compu-tations are described in Box 2.)

The hypothetical revenue effects of changes in tax law displayedin Table 16 differ from actual revenue estimates because they do nottake into account timing effects of changing the tax law. In particular,the 1986 act accelerated realizations from future years into 1986because the lower rates under pre-1986 law remained in effect be-tween the date of enactment and January 1, 1987; these additionalrealizations resulted in higher tax collections in fiscal year 1987.Table 16 instead displays the consequences for annual revenue, apartfrom immediate timing response, of the change in the structure of taxrates on capital gains implied by the realizations response estimatesin this paper.

The realizations responses estimated in all four equations showthat the 1986 act increased revenue from capital gains taxes and thata lowering of the top rate on capital gains to 15 percent would reducerevenue from capital gains taxes. The estimated revenue increasesfrom the 1986 act are, however, much lower than the increase that


TABLE 16. COMPARISON OF CAPITAL GAINS REVENUE UNDERPRE-1986 LAW WITH REVENUE UNDER CURRENT LAWAND UNDER A 15 PERCENT MAXIMUM RATE(In billions of dollars)

Behavioral Response

No Realizations Response

Equation 1 Response

Equation 2 Response

Equation 3 Response

Equation 4 Response

Current

No Realizations Response

Equation 1 Response

Equation 2 Response

Equation 3 Response

Equation 4 Response

TotalRevenues a/

Pre-1986 Law

39.0

Current Law

61.4

44.5

44.9

42.4

41.6

ChangePre-1986

Law

--

+ 22.4

+ 5.5

+ 5.8

+ 3.4

+ 2.5

fromCurrent

Law

-

-

-

-

-

-

Law, with 15 Percent MaximumRate on Capital Gains

35.3

37.1

37.1

37.5

37.7

-3.7

-1.9

-2.0

-1.5

-1.3

-26.1

-7.4

-7.8

-4.9

-3.9

SOURCE: Congressional Budget Office simulations using published data from the Internal RevenueService.

NOTE: In estimating the 15 percent maximum rate, it was assumed that the restored capital gainspreference would not be included in the base of the alternative minimum tax.

a. At 1988 levels, based on January 1987 CBO economic assumptions.


BOX 2REVENUE SIMULATIONS

Tax payments by individual taxpayers are simulated using a sample of 80,000tax returns for tax year 1984 produced by the Statistics of Income Division ofthe Internal Revenue Service. It is a stratified sample that over-representshigh-income returns; the separate observations are weighted to add up to theentire taxpaying population. The data were extrapolated to 1988, withdifferent types of income grown according to CBO economic assumptions.CBO has also developed a tax calculator that can be applied to the tax returndata in the sample to compute tax liability; the tax calculator can be modifiedto reflect actual or proposed changes in the tax law.

All of the revenue computations are at levels of income projected in taxyear 1988. In simulating the effects of changes in capital gains taxation, thefirst step is to compute marginal tax rates on the last dollar of long-termcapital gains for all taxpayers at levels of realizations projected under pre-1986 law. Marginal tax rates are computed under pre-1986 law, current law(after tax reform), and with a maximum tax rate on long-term gains of 15 per-cent. Then, the estimates of the realizations response reported in Table 10,combined with the changes in marginal tax rates on gains for each taxpayer,are used to compute new levels of realizations for all taxpayers for current lawand with a 15 percent maximum rate. The levels of realizations projectedunder pre-1986 law and those simulated for the two alternatives are then com-bined with other data on tax returns to compute taxable income and tax liabil-ity for each taxpayer under all three tax laws.

The simulation of the effects of changes in tax rates on realizations usedthe responses estimated in the equations for the entire sample in Table 10 in-stead of the separate responses for the top 1 percent and bottom 99 percent ofreturns shown in Tables 12 and 13. The reason for not using the separateresponses is that the estimates for the bottom 99 percent have a high standarderror and are therefore not too reliable statistically. Moreover, as noted in thetext, the estimates fail to reject the hypothesis that the responses are the samefor the bottom 99 percent and the top 1 percent. Even though a commonresponse is used for all taxpayers in the simulations, the revenue effects stilldepend on the structure of the tax rate change because the elasticity of reali-zations is larger at higher tax rates. This means that an increase in the aver-age marginal tax rate on gains among all taxpayers will have a smaller ad-verse effect on realizations if the percentage increase in marginal tax rates isrelatively larger among low-bracket taxpayers. Revenue from capital gainstaxes is maximized at a flat-rate tax on capital gains equal to the revenue-maximizing rate implied by the estimate of the realizations response.


would occur if there were no behavioral response. The no-behavioralresponse (or static) revenue gain is $22.4 billion; the net gain withbehavioral response varies from $2.5 billion (Equation 4) to $5.8billion (Equation 2). Thus, the realizations response is estimated tooffset between 75 percent and 90 percent of the static revenue pickupfrom tax reform.

Since the enactment of the Tax Reform Act of 1986, there has beensome discussion in the Congress of a proposal to reduce the maximumtax rate on long-term capital gains to 15 percent. Proponents of thisproposal have claimed that it would increase revenue because the in-crease in realizations would offset the direct loss from lower rates. 137The most likely realizations responses estimated in Equations 1-4,however, imply that a 15 percent maximum rate would reduce annualrevenues from capital gains taxes by between $4 billion and $8 billion,compared with current law. Revenues would be lower than under pre-1986 law by between $1 billion and $2 billion. Still, as noted above,the range of uncertainty in the coefficient estimates is large enoughthat one cannot reject the possibility that a 15 percent rate might in-crease revenues.

EVALUATION OF FINDINGS

Four major themes emerge from the data presented in Tables 10through 16. First, the statistical results confirm earlier research find-ings that higher marginal tax rates on capital gains lower realizationsof capital gains. The results are consistent with the widely held viewthat any overall revenue pickup from raising capital gains taxes isconsiderably less than the static pickup. They also support the gen-eral practice of taking account of behavioral responses to capital gainstaxes used in the official revenue estimates of the Joint Committee onTaxation and the U.S. Department of the Treasury. They also suggestthat tax rates on capital gains cannot be raised too high withoutultimately reducing revenue.

13. For a discussion of this proposal, and the views of supporters and opponents, see Lawrence J. Haas,"A 15PercentSolution?"MztiorcaJe/oumaHNovember7,1987).


Second, as shown in this chapter, historical evidence suggeststhat, in spite of the behavioral response, the increase in capital gainstax rates in the 1986 act most likely will increase revenue from capitalgains taxes, while a reduction of the top rate to 15 percent would mostlikely reduce revenue. The realizations response used in these esti-mates is qualitatively consistent with some earlier research, but issmaller than the response estimated in some of the studies that usecross-section data.

Third, the evidence of a significant realizations response is muchgreater for very-high-income taxpayers than for the remainder of thepopulation, although statistical tests fail to confirm a clear differencebetween the top 1 percent and the bottom 99 percent. Changes in theshare of gains realized by the top 1 percent of taxpayers in response topast tax rates on gains might be attributable to the fact that suchtaxpayers have experienced larger absolute changes in their marginaltax rates.

Fourth, it is important to emphasize that considerable uncer-tainty must be attached to the results of all statistical studies on therealizations of capital gains, including this one. The estimated effectof marginal tax rates on realizations is very sensitive to the selectionof other variables included in the equation as determinants of capitalgains. In addition, the standard errors around the estimated coeffi-cients in all equations, while allowing one to reject the hypothesis thatthere is no realizations effect of higher rates, allow for a wide range ofnegative responses. Moreover, relatively modest percentage changesin the realizations response translate into much larger percentagechanges in the net revenue effect of a tax change. As a result, stan-dard statistical tests do not permit rejection of the possibility that the1986 act reduced revenue from capital gains taxes or that a 15 percentrate would raise revenue. The latter possibility, however, is unlikely.

Finally, one important limitation of the analysis in this chapterneeds to be emphasized. The statistical results show only the effectthat changes in tax rates on capital gains would have on taxescollected from realized capital gains; they do not take into account allof the behavioral responses that could affect income tax revenues.Other components of the tax base, or even the size of the economy,would change as a result of changes in capital gains tax rates. Asnoted in Chapter n, these broader effects have not been quantified in


any of the statistical studies of the revenue effects of capital gainstaxation.

Aside from affecting realizations, changes in capital gains taxrates are most likely to alter the form in which income from capital isreceived, and its distribution among taxpayers. For example, a re-duction in capital gains tax rates that encouraged realizations couldalso encourage corporations to increase retained earnings at the ex-pense of lower dividend payouts and less debt financing. Both of thesechanges would lower individual income tax revenues because dividendand interest income is taxable, while unrealized appreciation is not.(Less debt financing would, however, raise corporate revenues.) Forassets such as real estate, lower capital gains tax rates that increasedthe turnover of assets would also increase depreciation write-offs,thereby lowering taxable business income, because the depreciablebasis of assets is increased when gains are realized. This effect alsowould tend to offset any positive revenue feedback from increasedrealizations.

On the other hand, a reduction in top rates on capital gains couldcause some high-bracket investors to substitute direct holdings of cor-porate equity for investments held through pension funds and invest-ments in owner-occupied housing and consumer durables. This substi-tution would increase revenue even if capital gains were tax-pre-ferred, because these alternative investments produce no taxable in-come.

Lower capital gains taxes could raise future taxable income by in-creasing total saving and economic growth, if private saving respond-ed positively to increases in after-tax rates of return. This effect wouldbe clearest if the lower taxes on capital gains were financed by in-creasing consumption taxes, resulting in a net decrease in the taxa-tion of saving. A shift from taxation of capital gains to taxation of in-terest and dividend income might only alter the form of saving andinvestment, but not the level. In any event, even if the response ofsaving to after-tax returns was positive, it would increase GNP andthe total tax base only gradually over time. Even then, the overalleffect on the tax base would probably be modest because capital gainstaxes are only a small part of the overall taxation of capital income.For example, one simulation study, which assumed a significantresponse of saving to after-tax returns, placed the increase in income


from the capital gains tax reduction in the 1978 act at much less than 1percent after 50 years. 147

Finally, it is important to emphasize that many factors other thanrevenue effects deserve consideration in deciding how to tax capitalgains. These considerations could be used to justify capital gains rateseither below or above the estimated revenue-maximizing rate. Lowertax rates on gains could increase saving and capital formation andchannel more resources into venture capital. In addition, a preferen-tial rate on gains provides a rough adjustment for the fact that aportion of nominal gains merely compensates taxpayers for thereduction in purchasing power of assets because of inflation. (Alterna-tively, one could directly eliminate the taxation of the inflationarycomponent of gains without introducing a preferential rate, by in-dexing the basis for changes in the general price level.) On the otherhand, keeping the rates on long-term gains, short-term capital gains,and ordinary income the same simplifies the tax system and promotesan efficient allocation of capital by equalizing tax rates amongdifferent types of financial instruments and real assets.

14. Office of the Secretary of the Treasury, Office of Tax Analysis, Capital Gains Tax Reductions, pp.97-147 and expecially p. 131.

APPENDIXES

APPENDIX A

ALTERNATIVE ESTIMATES OF THE

RESPONSE OF CAPITAL GAINS

REALIZATIONS TO TAX RATES

The estimates reported in Table 10 of Chapter IV are only a few of thepotential estimates that may be derived from equations that explainthe realizations of capital gains using marginal tax rates on gains andother variables. This appendix reports alternative specifications ofequations relating realizations to tax rates. These additional specifi-cations include minor modifications of the basic equations and the useof alternative functional forms.

The minor modifications of the basic equation test a number ofhypotheses. They examine whether it is possible to detect from thedata either lagged or anticipatory effects of tax changes on capitalgains realizations, and whether the realizations response is affectedby inclusion of selected other variables. The changes in functionalform are of two types. In the first, the consequences of changes in thefunctional form of the tax rate variable are examined. In the second,an alternative general functional form is estimated. The alternativegeneral functional form is a nonlinear equation that posits a multipli-cative relationship between realizations and estimated accrued gains,and that represents accrued gains as a linear combination of observedvariables that affect both the total value and the tax basis of capitalassets.

A concluding section examines in more detail the bias in the sepa-rate equations for the top 1 percent and bottom 99 percent of returnsthat results because induced realizations are included in the classifierused for splitting the sample. The procedure used for estimating thelikely consequences of this bias is explained.


MINOR MODIFICATIONS OF THE BASIC EQUATION

Timing Effects

Lagged Tax Effects. As noted in Chapter IV, the coefficients of themarginal tax rate variable in the equations in Table 10 will be biasedestimates of the long-run effect of tax rate changes on realizations ifthere are lagged responses to tax rates. If the long-run effect issmaller than the short-run effect, then the coefficient on the tax rateoverstates the absolute size of the long-run effect; if the long-run effectis larger than the short-run effect, the coefficient on the current taxrate understates the long-run effect.

One way of testing the bias is to include the tax rate of the previ-ous year in the equations in Table 10. As an illustration, Equation A2in Table A-l shows the effects of entering a lag term in Equation 2 ofTable 10. (For purposes of comparison with alternative specifica-tions, Equation 2 of Table 10 has been rewritten as Equation Al inTable A-l.)

The negative coefficient on the lag term means that a less thancomplete adjustment to tax rate changes occurs in the first year. Thelagged tax rate coefficient is, however, very small and not signifi-cantly different from zero. In comparison to the estimate in EquationAl that a one-percentage-point increase in tax rates reduces realiza-tions by 3.10 percent, Equation 2 implies that a one-percentage-pointincrease in tax rates reduces realizations by 3.00 percent in the firstyear and by 3.15 percent in the long run (the sum of the current andlagged coefficients). The implied long-run revenue-maximizing rate isonly slightly lower than in Equation Al-31.7 percent instead of 32.3percent.

Entering lagged variables into other equations (not reported here)gave the similar result that the coefficient on the lag term was notsignificantly different from zero. In some cases, the coefficient on thelag term was positive (indicating an overshooting in the first year); inother cases, it was negative (indicating, as above, less than full re-sponse in the first year). Thus, no time pattern of response has beenidentified.

APPENDIX A ALTERNATIVE ESTIMATES 75

The failure to find any effect with a single-period lag term shouldnot be regarded as a conclusive finding that the entire realizationsresponse occurs in the first year. More complex lagged relationshipsover a longer period of time could turn up either positive or negativelag effects. Because of the limited number of degrees of freedom in theequations, it is not feasible to test more complex relationships. Whatcan be said, however, is that no statistical evidence was found of alagged response of realizations to changes in the tax rate on capitalgains.!/

Anticipation of Future Tax Changes. When changes in tax rates areexpected to occur in a future year, the time pattern of realizations islikely to be affected. If tax rates on capital gains are expected to in-crease, taxpayers are likely to accelerate realizations in order to paytaxes at the lower rate; in contrast, if rates are expected to decline,taxpayers have an incentive to delay realizations.

Expected future tax rate changes are of two types. Sometimesthey are already built into current law, as when a tax law change hasbeen enacted with a future effective date. An example is the TaxReform Act of 1986, signed by the President on October 22, 1986,which eliminated the capital gains deduction effective on January 1,1987. A second and more ambiguous situation is when taxpayersexpect a change that has not yet been enacted into law. Because of thelong time it takes to enact tax laws, investors frequently anticipatechanges in advance of the legislation. (Sometimes changes are maderetroactive in recognition of this possibility. For example, thereduction in the top capital gains rate to 20 percent in 1981 waseffective June 20, 1981, even though the tax law change was notenacted until August.) It is not clear, however, at exactly what stageof the legislative process these expectations become strong enough toaffect behavior.

In contrast, the Treasury Department, using a sample for the years 1954-1982 instead of 1954-1985, did find evidence that realizations are related positively to the previous year's tax rate. Thisis interpreted as meaning that the long-run realizations response is smaller (in absolute terms)than the first-year effect. See Office of the Secretary of the Treasury, Office of Tax Analysis, Reportto Congress on the Capital Gains Tax Reductions of 1978 (September 1985), pp 175-176.


TABLE A-l. ALTERNATIVE EQUATIONS EXPLAINING CAPITALGAINS REALIZATIONS FOR ALL TAXPAYERS,1954-1985Dependent Variable is log(LTG)

Equation andDescription Constant

Al

A2

A3

A4

A5

A6

Basic(Equation 2from Table 10) -6.82

Adjustment toPast Tax Change -6.84

Anticipation ofComing Tax Change -6.80

AutocorrelationCorrection -7.66

Holding Period(In months) -6.98

Compliance(0 before 1983) -7.33

logPRICE

1.17(5.75)

1.17(5.45)

1.17(5.60)

1.17(5.05)

1.15(5.16)

1.11(4.96)

logRCOEQ

0.501(4.19)

0.497(3.66)

0.500(3.93)

0.565(4.64)

0.506(4.10)

0.488(4.01)

logRGNP

1.02(3.05)

1.03(2.88)

1.02(2.99)

1.07(2.76)

1.03(3.00)

1.08(3.13)

dlogRGNP

2.06(2.57)

2.06(2.51)

2.07(2.45)

2.30(2.90)

2.04(2.46)

1.79(2.03)


NOTES: t-statistics are in parentheses. Equation A4 is estimated from 1955 instead of 1954 because ofthe first-order autoregressive correction (Cochran-Orcutt technique). The Durbin-Watsonstatistic is not appropriate (n.a.) in equation A4.

Variables appearing in all equations are:

log( LTG) = logarithm of net long-term gains, in excess of net short-term losses;log(PRICE) = logarithm of the GNP deflator;log(RCOEQ) = logarithm of corporate equity held by households in constant dollars;

(Continued)


TABLE A-l. Continued

MTRNewTerm R2

RegressionStatistics

S.E.E. D.W.

Revenue-Maximizing

Tax Rate

-0.0310(-2.47)

-0.0300(-1.61)

-0.0311(-2.40)

-0.0319(-2.28)

-0.0309(-2.41)

-0.0267(-1.94)

•"

MTR(-l)-0.00149

(0.07)

dMTR( + l)0.00490(0.05)

AR(1)0.344(1.63)

HOLDING0.00300(0.23)

COMPLY0.0845(0.78)

.988 .102 1.40 32.3

.988 .104 1.42 31.7

.988 .104 1.40 32.2

.988 .100 n.a. 31.4

.988 .104 1.41 32.4

.988 .103 1.47 37.4

NOTES: (Continued)

log(RGNP)dlog(RGNP)MTR

= logarithm of GNP in constant dollars;= change from the preceding year in log(RGNP); and= weighted average marginal tax rate on capital gains.

Variables appearing as new terms in single equations are:

MTR(-l) = MTR from the preceding year (the 1954 value was used for 1953);dMTR(-t-l) = legislated change in MTR in the coming year;ARID = first-order autoregressive term;HOLDING = number of months asset must be held to qualify for long-term capital gains

treatment; andCOMPLY = one in years of broker reporting on stock sales (1983-1985), zero otherwise.


An attempt was made to test statistically for the effect of anti-cipated changes by including a variable for the change in the marginaltax rate on gains in the following year for years in which changesaffecting marginal tax rates on capital gains were already enactedinto law, but not in effect. There were only a few years in the samplein which this situation occurred:

o The 1964 act phased in lower marginal income tax rates overa two-year period. This change affected capital gains taxrates for taxpayers below the 50 percent bracket. (The 50percent deduction for long-term gains and the 25 percentmaximum rate on gains were unaffected by the act.) Thus,in 1964, the anticipated gains rate for 1965 was lower thanthe current rate for many taxpayers.

o The 1969 act eliminated the 25 percent maximum rate onlong-term gains with a three-year phase-in between 1970and 1972. Thus, in 1969, 1970, and 1971, the anticipatedfuture year's rate was higher than the current rate.

o The 1978 act increased the capital gains deduction to 60percent, removed the capital gains preference from the add-on minimum tax, and eliminated the "poisoning" of themaximum tax on earned income by capital gains. The in-crease in the deduction was effective October 31, 1978, butthe minimum tax and maximum tax provisions were delayeduntil January 1, 1979. This delay made the anticipatedfuture rate in 1979 slightly lower than the rate in 1978.

o The 1981 act lowered the maximum rate on capital gains to20 percent, effective June 20, 1981, but phased in the mar-ginal tax rate cuts on ordinary income over a three-yearperiod. Taxpayers below the 50 percent marginal ratebracket thus could expect declines in capital gains tax ratesin 1982,1983, and 1984. This makes the anticipated futuretax change variable negative for the years 1981-1983, al-though only slightly negative on average because the high-est-income taxpayers received their entire capital gains taxrate cut in 1981.


Equation A3 in Table A-l shows the effect of including an antici-pated tax rate term in the equation. The coefficient is positive,showing that an enacted future tax rate increase raises current reali-zations, but it is not significantly different from zero. Inclusion of thefuture tax rate change in the equation has virtually no effect on theestimated response to the current rate.

The failure to identify an effect of legislated future changes oncurrent realizations reflects the fact that, before 1986, delays in theeffective date of changes in tax rates on capital gains were not thatimportant. Further, within-year changes in the timing of realizationscannot be captured by the equations because only annual data onrealizations are available. Thus, for example, in 1978 Congressionalsupport for lowering the capital gains rate was evident quite early inthe year. The 1978 act became law in October, and the effective dateof the increase in the capital gains deduction was October 31. It isquite possible that taxpayers delayed gains realizations in 1978 untilNovember to take advantage of the lower tax rate, but the data cannotcapture this effect.

The effective date of the capital gains increase in the 1986 actprovided a substantial incentive for taxpayers to accelerate realiza-tions from 1987 and future years into 1986. Data that can be used toestimate the acceleration effect of the 1986 act are not yet available.

Correction for Autocorrelation of Residuals. The estimated Durbin-Watson (D.W.) statistic in Equation Al suggests that there might beserial correlation of the residuals. Serial correlation does not result inbiased estimates of coefficients, but could result in understatement ofstandard errors.

Equation A4 duplicates Equation Al, with a first-order auto-correlation term. The coefficient of AR(1) indicates a positive serialcorrelation of the residuals, but the coefficient is not significantly dif-ferent from zero at the 5 percent level. The use of the autocorrelationcorrection has only a very slight effect on the other coefficients of theequation. The estimate of the reduction in capital gains realizationsfrom a one-percentage-point increase in marginal tax rates on gainsincreases slightly from 3.10 percent to 3.1^ percent, and the impliedrevenue-maximizing tax rate declines from 32.3 percent to 31.4

82-667 0 - 88 - A


percent. The tax rate effect continues to be significantly differentfrom zero.

Other Variables

Holding Period. An increase in the holding period for determiningwhether a gain is long-term can be expected to reduce long-term gains.This effect takes place because gains on assets held longer than the oldholding period, but not as long as the new holding period, would be re-classified from long-term to short-term.

The holding period changed in three years of the sample period.In the 1976 act, it was increased from six months to nine months in1977 and twelve months in 1978 and succeeding years. In the 1984act, it was reduced to six months for gains realized in 1985-1987.

The holding-period effect is not expected to be large. The over-whelming share of realized capital gains are long-term even in yearswhen the holding period is one year. For example, in 1984 netlong-term gains (in excess of net short-term losses) were $134.1 bil-lion; net short-term gains were only $4.8 billion.

The effect of the holding period on long-term realizations is testedby adding to the basic equation a variable for the number of months anasset must be held to qualify as long-term. The result is shown inEquation A5 of Table A-l. The coefficient of the variable HOLDING issmall, statistically insignificant, and positive instead of negative. Theaddition of the holding-period variable does not significantly alter thecoefficients of the other variables or the statistical fit of the equation.

Reporting Requirements. The Tax Equity and Fiscal ResponsibilityAct of 1982 introduced a number of new provisions to improve com-pliance with the income tax. Among these new provisions was arequirement that brokers report sales of assets to the Internal Reve-nue Service and supply information returns to taxpayers. Reportingwas first required for assets sold after July 1,1983.

The effect of the new reporting requirements is tested by in-cluding a "dummy" variable for compliance that is assigned the valuezero for years before 1983 and the value one for 1983 through 1985.


This variable, called COMPLY, though meant to test for the effect ofcompliance provisions, will capture any structural shift in the equa-tion for capital gains realizations for years beginning in 1983.

Equation A6 in Table A-l shows the effect of adding the variableCOMPLY to equation Al. The coefficient of COMPLY implies thatsome unspecified structural change, which could be interpreted as theeffect of the compliance provisions, raised capital gains realizations by8.45 percent above the level otherwise predicted by the equation. Thestandard error of the coefficient of COMPLY, however, is almost aslarge as the coefficient, indicating that the coefficient is not signifi-cantly different from zero. Including COMPLY in the equation resultsin a smaller estimate of the tax rate effect; the decline in realizationsassociated with a one-percentage-point increase in marginal rates is2.67 percent, compared with 3.10 percent in Equation Al. The esti-mate of the revenue-maximizing tax rate on capital gains is 37.4 per-cent, instead of 32.3 percent.

The estimated 8.45 percent increase in reporting attributed to thecompliance provisions of the 1982 act is about the most that could beexpected from the provisions. In a recent study of the effects of thecompliance provisions, Auten concludes that even complete reportingof gains on stocks and bonds would not raise total realizations by morethan about 7.5 percent.2/ Underreporting of capital gains has beenfound to be between 15 percent and 25 percent in most years checked,while gains on corporate stocks and bonds account for about 30 percentof gains reported.3/ Thus if underreporting on stocks and bonds is ascommon as on other assets, full reporting of gains on stocks and bondswould raise total realizations by 7.5 percent (30 percent of the 25 per-cent underreported). This is probably a maximum figure becausegains on stocks appear to be less subject to underreporting than otherassets. Of course, the reporting requirements could serve to remind

2. Gerald E. Auten, The Effect of Security Transactions Reporting Requirements on the Reporting ofCapital Gains, Report Prepared for the Research Division of the Internal Revenue Service,September 22,1986, pp. 5-8.

3. Auten refers to a study for the Internal Revenue Service by ICF Incorporated in whichunderreporting of capital gains was estimated to range between 16 percent and 26 percent duringthe 1972 to 1983 period. (Gerald E. Auten, The Effect of Security Transactions ReportingRequirements, Table 4). Furthermore, the Internal Revenue Service has also estimated under-reporting to be between 15 percent and 25 percent in four of six years studied between 1965 and1982. See James M. Poterba, "Tax Evasion and Capital Gains Taxation," The American EconomicReview, vol. 77 (May 1987), pp. 235-236.


taxpayers to report all gains, and in this way reporting could improveby more than the 7.5 percent possible from stocks and bonds alone.

In his study, Auten uses a similar approach to that of equation A6and estimates an even larger upward shift in capital gains in and after1983. Because the reporting requirements were only effective for partof 1983, Auten experiments with different degrees of response in 1983and 1984 (his data ended in 1984). Auten finds compliance effects of10 percent to 20 percent, depending on the year and the specifica-tion.4/ Because his estimated effects are greater than could be ex-pected from improved reporting alone, Auten concludes that otherevents in 1983 and 1984 probably contributed to the estimated effectsof compliance.

Use of Wealth Instead of Income Variables. The equations reported inChapter IV and in Table A-l of this appendix explain capital gainsrealizations as a function of marginal tax rates on capital gains, thevalue of corporate equity, GNP, and the change in GNP. In a recentpaper, Lindsey estimated equations in which capital gains, in constantdollars, were represented as depending on marginal tax rates andmeasures of wealth in different types of assets (also in constant dol-lars) .57 Lindsey's sample of observations is a pooled cross-section timeseries for six AGI groups for the years 1965-1982. His estimates of thetax rate effect therefore depend both on how differences in realizationsamong taxpayer groups in the same year relate statistically to dif-ferences in marginal tax rates among groups and on how differencesover time in realizations relate statistically to changes over time inmarginal tax rates. In comparison, this study examines only time-series relationships.

Table A-2 reports time-series equations relating capital gainsrealizations to marginal tax rates and the measures of wealth used byLindsey. The data for the years 1965-1982 on wealth and changes in

4. Gerald E. Auten, The Effect of Security Transactions Reporting Requirements, pp. 18-20 andTable 7.

5. See Lawrence B. Lindsey, Capital Gains: Rates, Realizations and Revenues, Working Paper No.1893 (National Bureau of Economic Research, Inc., April 1986).


TABLE A-2. EQUATIONS EXPLAINING CAPITAL GAINSREALIZATIONS USING WEALTH INSTEADOF INCOME VARIABLES (All taxpayers)Dependent variable: log(LTG)

Equation andDescription Variables and t-statistics R2

Regression Revenue-Statistics Maximizing

S.E.E. D.W. Tax Rate

CBO VariablesCon- log log dlogstant RCOEQ RGNP RGNP MTR

A7 Equation 4fromTable 10

-8.42 0.425(5.42)

1.30(20.2)

1.98(2.49)

-0.0388(-4.51)

Wealth VariablesCon-stant

logTRD

logNTRD

logREV MTR

.951 .102 1.42 25.8

A8 Lindsey TaxRate 1965- -6.111982

2.28 -1.17 0.0415 -0.00012 .694 .120 1.73 Over 100(2.95) (-2.05) (1.02) (-0.01)

A9 CBO TaxRate 1965- -5.71 2.14 -1.07 0.0460 -0.00454 .696 .120 1.69 Over 1001982 (2.54) (-1.68) (1.09) (-0.24)

A10 CBO TaxRate 1954-1982

-10.8 2.10 -0.391 0.0395 0.00130 .872 .142 1.02 Over 100(2.60) (-0.65) (0.86) (0.06)

All CBO TaxRate 1954- -8.49 1.12 0.392 0.0741 -0.0254 .904 .142 1.01 39.41985 (1.97) (1.02) (1.81) (-2.14)


NOTES: t-statistics are in parentheses.

log(RLTG) = logarithm of net long-term gains in excess of net short-term losses, inconstant dollars;

log(RCOEQ) = logarithm of corporate equity held by households in constant dollars;log(RGNP) = logarithm of GNP in constant dollars;dlog(RGNP) = change from the preceding year in log(RGNP);MTR = weighted average marginal tax rate on capital gains;log(TRD) = logarithm of tradable wealth in constant dollars;log(NTRD) = logarithm of nontradable wealth in constant dollars; andlog(REV) = logarithm of real revaluations in tradable wealth.


asset values are constructed using the sources and methods describedby Lindsey, except that they are not allocated among income classes.6/

Lindsey constructed an estimate of the marginal tax rate oncapital gains for each AGI group for the sample period in his studythat is slightly different from the estimate of the marginal tax rate ongains for each AGI group constructed by CBO. In equations for theyears 1965-1982 in Table A-2, both measures of marginal tax rates byAGI group are aggregated into a single average marginal tax rate onlong-term gains for the entire sample.?/ For years before 1965 andafter 1982, only the CBO marginal tax rate measure is available.

The equations estimated by Lindsey used realized gains in con-stant dollars, not current dollars, as the dependent variable. There-fore, for purposes of comparing CBO results with the exact specifica-tion of wealth measures used by Lindsey, Equation A7 in Table A-2repeats Equation 4 from Table 10, which also uses long-term gains inconstant dollars as the dependent variable. Lindsey groups all house-hold assets into two variables, the value of tradable wealth and thevalue of nontradable wealth. He also includes the revaluation of trad-able wealth in the variables explaining gains. All measures are inconstant dollars.

Equation A8 is estimated for the years 1965-1982, using the aver-age marginal tax rate measure constructed from Lindsey's tax rates.The estimate of the tax rate effect from this time-series equation isvery small and not significantly different from zero. The coefficientsof the wealth variables have the same sign as estimated by Lindsey,but the coefficient of the current year revaluation is not significantlydifferent from zero. Equation A9 is the same as Equation A8, exceptthat it uses CBO's marginal tax rate measure. Substitution of theCBO for the Lindsey tax rate measure affects the equation only slight-ly; the estimated marginal tax rate effect becomes larger, but is stillsmall and not significantly different from zero.

6. As in the equations used in this paper, the wealth data used by Lindsey are those reported in Boardof Governors of the Federal Reserve System, Balance Sheets for the U.S.Economy.

1. The marginal tax rate on gains for the entire population is computed as a weighted average of themarginal tax rate for each AGI group. The weights are "predicted" gains in each AGI class, with"predicted" gains estimated as described in footnote 11 of Chapter III.


Equations A10 and All use the Lindsey wealth measures (andCBO's marginal tax rate measure) for the years 1954-1982 and 1954-1985, respectively. Expansion of the sample back to 1954 has littleeffect on the estimate of the tax rate effect, but extension of the sampleforward to 1985 makes the estimated effect of tax rates on realizationsnegative and statistically signficant. A one-percentage-point increasein marginal tax rates on gains lowers realized long-term gains (in con-stant dollars) by 2.54 percent, compared with a 3.88 percent reductionin the CBO specification.

The results reported in Table A-2 differ considerably from thefindings by Lindsey because they are estimated over a different timeperiod and because the taxpayers in different income groups are notused as separate units of observation.8/ They are shown only for thepurpose of illustrating the consequences of a specification of the equa-tion in Table 10 that uses wealth instead of income variables to ex-plain changes in gains.

OTHER FUNCTIONAL FORMS

The equations reported above are similar to those in Table 10 of thetext in one way—they all use the same functional form. The relation-ship between gains and nontax variables is represented in logarithmicform, while the relationship between gains and tax rates is log-linear.Because the functional form of the relationship could affect the re-sults, estimates were made using equations with different functionalforms. This section also shows the consequences of using a more ex-plicit measure of accrued gains to explain realizations of gains.

Changes in the Tax Rate Variable

In Chapter IV, the estimated equation is of the form:

log (LTG) = a0 + ai*log(X) + a2*MTR,

8. Lindsey's findings are summarized in Chapter II.


where LTG = realized long-term gains, MTR = the marginal taxrate on gains (in percents), and X = a vector of other variables be-lieved to influence gains. The specification of a log-linear relationshipbetween realized gains and the marginal tax rate on gains impliesthat the elasticity of realizations with respect to the tax rate becomeslarger (in absolute terms) as the tax rate increases. For absolutevalues of the tax rate elasticity that are less than one, revenuechanges in the same direction as tax rates; for elasticity values greaterthan one, the revenue effect is opposite in sign to the tax rate change.The revenue-maximizing rate, which occurs at an elasticity equal to-1.0, can be easily calculated as MTR* = -l/a2-

An alternative functional form that also permits a simple calcula-tion of the marginal tax rate is a log-log form using the after-tax gainper dollar of pretax gain as the independent variable. Algebraically,this is represented as:

log (LTG) = a0 + ai*log(X) + a2*log(100-MTR).

The value a2 is now expected to be positive; that is, as the after-taxproceeds per dollar of pretax gain increase, realized gains will also in-crease.

As with the functional form used in Chapter IV, the form that islogarithmic in the after-tax gain implies that the elasticity withrespect to the marginal tax rate increases (in absolute value) as thetax rate increases. It also implies a revenue-maximizing tax rate, cal-culated as:

T* = 11(1 + a2).

Equations A12 and A13 in Table A-3 compare the two functionalforms for the tax variable. Equation A12 repeats the equation with asemilog form reported as Equation 2 in Table 10 and Equation Al inTable A-l. Equation A13 is the same as Equation A12, except that thetax rate variable is the logarithm of 100 minus the marginal tax rate.The coefficients of the nontax variables and the fit of the equation arevirtually unchanged by the substitution of the new tax rate variable.The implied revenue-maximizing tax rate declines slightly from 32.3percent to 28.3 percent; the lower revenue-maximizing rate results


from a greater curvature outside of the sample range (discussedbelow).

Equation A14 in Table A-3 reports the result of a logarithmic spe-cification between realizations and marginal tax rates. In EquationA14, there is no revenue-maximizing rate because the functional formassumes a constant elasticity. The estimated elasticity of -0.554 im-plies that revenue increases as marginal tax rates are raised, for alltax rates.

Finally, Equation A15 shows the results of entering the tax ratewith both a linear and a squared term (quadratic form). In this form,the implied curvature of the revenue function is much greater than inthe others, and the revenue-maximizing rate is only 21.4 percent. Theseparate coefficients of MTR and MTR2, however, are not statisticallysignificant, indicating that the estimates of the exact curvature of thisrelationship are not reliable.

Figure A-l shows the relationship between revenues and mar-ginal tax rates under the four alternative specifications of the tax ratevariable. Between average marginal tax rates of 13 percent and 23percent—the limits of the average marginal tax rates for all returns inthe sample period—the shapes of all four estimated relationships arepractically identical. The four functional forms do, however, implydifferent relationships between revenue and tax rates outside thesample range. The logarithmic form has the least curvature, implyingthat revenues will continue to increase no matter how high marginaltax rates are increased. With all the other functions, revenues even-tually decline as the marginal tax rate is increased. The revenuecurves for the log-linear and after-tax return specifications begin tomove farther apart for rates above 30 percent; both curves showrevenue eventually declining with higher marginal tax rates, but thedecline is greater with the after-tax return specification. The quad-ratic specification has the greatest curvature, with revenue peaking at21 percent and declining sharply at higher rates.

Nonlinear Equations with Accrued Gains

As discussed in Chapter IV, accrued gains depend on both the value ofassets and their cost basis. Increases in the value of assets increase


accrued gains; increases in basis reduce accrued gains. Data on thecost basis of assets are not available.

While basis cannot be observed directly, it does depend on othervariables that can be measured. Realization of gains in previous yearsreduces current asset basis; transfer of assets with gains at death alsoreduces basis; tax depreciation deductions in excess of economic depre-ciation increase basis. Past realizations of capital gains can be ob-served directly (although unreported gains, which also increase basisif reinvested, are unobserved); gains at death should vary directlywith the number of deaths in the population; and excess depreciationdeductions claimed by individuals can be estimated from reported

TABLE A-3. EQUATIONS EXPLAINING CAPITAL GAINSREALIZATIONS USING ALTERNATIVE TAX RATESPECIFICATIONS (All taxpayers, 1954-1985)Dependent variable: log(LTG)

Equation andDescription Constant

logPRICE

logRCOEQ

logRGNP

dlogRGNP

A12 Basic(Equation 2

A13

A14

A15

from Table 10)

After-tax Rateof Return

LogarithmicTax Rate

QuadraticTax Rate

-6.82 1.17(5.75)

-18.5 1.18(5.80)

-5.83 1.16(5.54)

-7.31 1.23(5.36)

0.501(4.19)

0.501(4.19)

0.504(4.18)

0.504(4.15)

1.02(3.05)

1.02(3.06)

1.03(3.01)

0.971(2.77)

2.06(2.57)

2.08(2.60)

1.98(2.45)

2.34(2.46)


NOTES: t-statistics are in parentheses.

log(LTG) = logarithm of net long-term gains in excess of net short-term losses;log(PRICE) = logarithm of GNP deflator;

(Continued)


data in the National Income and Product Accounts. Thus, using in-complete data, one can express accrued gains as a function of variablescorrelated with wealth in capital assets and variables correlated withbasis.

Table A-4 reports an equation in which accrued gains arerepresented as a linear combination of other variables. The equationfor realized gains is of the form:

(A16) log (LTG) = a0 + ai*log(PRICE) + a2*MTR + a3*log(RAG)

TABLE A-3. Continued

MTR

-0.0310(-2.47)

NewTerm R2

.988

log(lOO-MTR)2.53 .988

(2.48)

0.0680(0.38)

NOTES:

log(MTR)-0.554(-2.41)

MTR2

-0.00268(-0.56)

(Continued)

log(RCOEQ)log(RGNP)dlog(RGNP)MTR

.988

.988

RegressionStatistics

S.E.E. D.W.

.102 1.40

.102 1.40

.103 1.42

.104 1.37

Revenue-Maximizing

Tax Rate

32.3

28.3

None

21.4

logarithm of corporate equity held by households in constant dollars;logarithm of GNP in constant dollars;change from the preceding year in log(RGNP); andweighted average marginal tax rate on long-term capital gains.


where LTG = realized long-term gains, PRICE = the price level,MTR = the marginal tax rate on long-term gains, and RAG = accruedgains in constant dollars. Accrued gains in constant dollars are repre-sented by:

(A17) RAG = bi*CORP + b2*NCORP + b3*GNS + b4*DEATH

Figure A-1.Relationship of Revenue from Capital Gains Taxes to the MarginalTax Rate on Gains for Four Specifications of the Tax Rate Variable

Revenue Index = 100 at 20 percent tax rate.160

140 -

120 -

100 -

40 -

20

0 fi I I I I I I I I I I I I I I I I I I I I I I I I I I I I I | I I I I I I I I I I I I I I I I I I

0 5 10 15 20 25 30 35 40 45 50

Marginal Tax Rate

SOURCES: Congressional Budget Off ice.


TABLE A-4. EQUATION EXPLAINING CAPITAL GAINS REALIZA-TIONS USING CHANGES IN ASSET VALUES AND BASIS(All taxpayers, 1954-1985)Dependent variable: log(LTG)

Variable

constant, GI

log(PRICE), ai

MTR, a2

Accrued gains, a3

NCORP, c2

GNS,c3

DEATH,c4

R2 = .982 S.E.E. = .126

Coefficient

-2.22

1.27

-0.0205

0.926

1.19

2.78

-74.2

D.W. = 1.43 T* = 48.7

t-statistic

-0.75

3.72

-1.34

2.85

1.73

1.50

-1.60


NOTES: log(LTG)log( PRICE)MTRAccrued gainsNCORP

GNSDEATHT*

logarithm of net long-term gains in excess of net short-term losses;logarithm of GNP deflator;weighted average marginal tax rate on long-term capital gains;defined in equation A17 of the text;sum of past revaluations of noncorporate equity plus the differencebetween tax and economic depreciation on noncorporate capital, inconstant dollars;sum of past realized gains in constant dollars;sum of deaths in previous years; andrevenue-maximizing tax rate.

where CORP = the sum of past revaluations of corporate equity,NCORP — the sum of past revaluations of noncorporate equity plusthe difference between tax and economic depreciation on noncorporatecapital, GNS = the sum of past gains, and DEATH = the sum ofdeaths in prior years.9/

Combining A16 and A17 gives the estimating equation forrealized gains:

9. All the sums of past variables are cumulated from 1947 except for past realizations of gains, whichare unavailable before 1954. All dollar amounts are assigned a starting value in the year prior tocumulation. The starting value is selected so as to make the calculated values consistent with dataon the ratio of gains to sales prices of assets in 1977. This ratio was calculated by the Joint TaxCommittee using IRS data.


(A18) log (LTG) = ci + ai*log(PRICE) + a2*MTR +a3*log(CORP + c2*NCORP + c3*GNS + c4*DEATH)

where ci = ao + a3*log(bi), C2 = b2/t>i, cs = bs/bi, andc4 = b4/bi.

Equation A18 is estimated by a nonlinear technique. The result isshown in Table A-4. The estimated coefficient of the marginal tax rateon gains continues to be negative, but is smaller than in the equationsreported in Table 10 of the text and is not statistically different fromzero. The implied revenue-maximizing rate is 48.7 percent, muchhigher than in any of the equations estimated using simple wealthand income measures.

Four of the other five coefficients in the equation have theexpected sign. The coefficient of the price term implies that a doublingof the price level raises nominal realizations by 127 percent; the esti-mated elasticity of nominal gains with respect to the price level issignificantly different from zero, but not significantly different fromone. The coefficient of accrued gains is also close to one, implying aunit elastic response of realizations to accruals. Two of the threeterms used to explain accrued gains have the expected sign. Pastrevaluations of noncorporate equity, net of excess tax depreciation, arepositively related to realized gains. An increase in deaths, whichlowers accrued gains by reducing basis, is found to be negatively re-lated to realizations. The coefficient on past realizations, however,does not have the hypothesized sign. Even though past realizationsshould increase basis, thereby reducing accrued gains, they are esti-mated to be positively related to current realized gains.

BIAS RESULTING FROM THE CLASSIFICATIONOF TAXPAYERS BY AGI

As discussed in Chapter IV, classifying taxpayers by AGI can result inbias in estimating separate time-series equations explaining gains fordifferent groups in the population. This bias results because AGIincludes realized gains. In years when gains increase relative to othersources of income, taxpayers whose gains account for a large share oftheir AGI will make up a greater than normal share of returns in thetop 1 percent. Thus, the increase in gains in the top 1 percent of re-


turns ranked by AGI will be greater than the increase in gains for thetop 1 percent of a fixed population of taxpayers. If the increase inrealized gains is induced by a decline in tax rates on capital gains, thetax effect on gains for the top 1 percent of returns ranked by AGI willbe overstated, and the tax effect on gains for the bottom 99 percentunderstated, compared with the induced tax effects that would be esti-mated for taxpayer groups whose composition does not change.

In the equations for separate groups in the population reported inTables 12 and 13 (in Chapter IV), returns must be classified by AGIbecause that is how published data are reported for the years 1954through 1985. For seven years of the sample period, however, samplesof individual tax returns were available that could be used to developrankings of taxpayers by other classifiers. For years in which thesedata were available, returns may be classified by AGI less gains, aprocedure that eliminates the bias that occurs when changes in gainsalter the composition of returns in the top 1 percent.

As a test of how the bias influenced the results for the entire timeperiod, an equation was estimated relating the shares of gains in thetop 1 percent under the two classifiers for the seven years in whichsamples of individual tax returns were available. The estimatedequation is of the form:

(A19) D = a + b*R

where D = the difference between the percentage of gains realizedby the top 1 percent classified by AGI and classified by AGI less gains,and R = realized gains as a share of AGI. The coefficient of b is ex-pected to be positive, reflecting that an increase in realized gains forthe entire population increases the extent to which gains in the top 1percent classified by AGI exceed gains in the top 1 percent classifiedby a measure (AGI less gains) that is unaffected by induced gains.

In Equation A19, the coefficient b is estimated to be 0.38, with astandard error of 0.28. The positive coefficient on b indicates a bias inthe direction hypothesized, but the standard error with only sevenobservations is so large that the coefficient is not statistically differentfrom zero.


The estimated equation was used to develop corrected measures ofgains for the top 1 percent and bottom 99 percent of the population.When the corrected measures are inserted in equation 1.01 in Table 12and in equation 1.99 in Table 13, the estimates of the tax response areonly slightly altered. The tax rate coefficient—that is, the estimatedpercentage change in realizations per percentage-point change in themarginal tax rate on gains—is reduced from -.0323 to -.0316 for the top1 percent of returns by using the corrected measure of gains. The taxcoefficient is increased from -.0329 to -.0348 for the bottom 99 percentof returns. These small changes in the coefficients suggest that thebias introduced from classifying the population by AGI, though in theexpected direction, is apparently small.

CONCLUSIONS

This appendix has reviewed a wide range of alternative ways ofspecifying equations for the realization of capital gains. Many of thealternative specifications do not significantly alter the results re-ported in Chapter IV. Lagged and anticipatory responses to tax ratechanges could not be identified from the sample, and equations incor-porating them did not alter the estimate of the realizations response.Changes in the functional form for the tax variable also do not havemuch effect on the results, unless they are used to project the effects oftax rate changes that are far outside the range of past historical ex-perience. Changes in the holding period and in compliance require-ments do not appear as significant determinants of realizations ofgains within the sample period. Finally, the bias in estimating theresponse for separate groups of taxpayers that results from using AGIas a classifier appears to be small.

The results are affected, however, by the choice of income andwealth variables used to represent the potential for realizing gains. Inthe specifications reported in this appendix, the realizations responseis greatest when potential gains are represented by including bothmeasures of income and wealth in corporate equities in the equation.When measures of wealth in other assets are substituted for the in-come measure, the estimate of the realizations response declines.When a proxy variable for accrued gains is constructed by includingvariables that contribute to both wealth and basis, the estimated


realizations response falls even more. Other combinations of vari-ables that were not tested could result in larger or smaller estimates ofthe realizations response.

The sensitivity of the estimate of the realizations response to thespecification of other variables included in the equation adds to theuncertainty of the results. This uncertainty cannot be resolved be-cause, even if all possible combinations of variables were tested, itcould still be debated which one represented the correct specification.

APPENDIX B

METHODOLOGICAL ISSUES IN STUDIES

OF THE REALIZATIONS RESPONSE

As shown in Chapter n, the results of studies of the realizationsresponse to tax rates on capital gains differ between those using cross-section and those using time-series approaches. The cross-sectionstudies tend to show a greater responsiveness of realizations to taxrates. This appendix briefly reviews the strengths and weaknesses ofeach.!,/ The review suggests that, while each has much to contribute,the cross-section approach is more likely to overstate the responsive-ness of realizations to tax rates. The limitations of the time-seriesapproach, on the other hand, can lead to a wide range of uncertaintyabout the correct response but do not clearly point toward over- orunderstatement.

CROSS-SECTION AND TIME-SERIES APPROACHES

Cross-section studies compare realizations among taxpayers with dif-ferent tax rates in a single year. Time-series studies compare realiza-tions for all taxpayers as tax rates change over time. Cross-sectionstudies may employ limited data from other years, as in calculatingexpected tax rates for the year in question, and time-series studiesmay disaggregate to subgroups of the population. A third alternativeis to combine time-series and cross-section analysis by using as sepa-rate observations data for individual taxpayers in multiple years. Taxdata for the same taxpayer in many years are generally unavailable,but groups of taxpayers classified by AGI have been used in one study

For other discussions of methodology, see Jane G. Gravelle, "A Proposal for Raising Revenue byReducing Capital Gains Taxes?" Congressional Research Service Report (June 30, 1987); Eric W.Cook and John F. O'Hare, "Issues Relating to the Taxation of Capital Gains;" National TaxJournal, vol. 40 (September 1987); and Office of the Secretary of the Treasury, Office of TaxAnalysis, Report to Congress on the Capital Gains Tax Reductions of 1978 (September 1985), pp.158-162.


by Lindsey.2/ He subdivided taxpayers into six income categories ineach of the years from 1965 to 1982.

ISSUES IN CROSS-SECTION STUDIES

Advantages of Cross-Section Studies

The advantages of cross-section studies obtain from their use of datafrom individual tax returns. First, the data from a single tax returnreflect the situation of a single behavioral unit-the single taxpayer ora family. Thus the level of realizations can be directly matched to thattaxpayer's tax rates, income from different sources, and family status.Second, thousands of records are available and multicollinearity tendsto be low. As a result, stable coefficients with low standard errors canbe estimated even though many determinants of gains are included.

Problems Unique to Cross-Section Studies

The major problems inherent in cross-section studies are the depen-dence of tax rates on taxpayer behavior, the commingling of responsesto temporary and permanent changes in tax rates, and the limitationsof data provided on tax returns.

Dependence of Marginal Tax Rates on Taxpayer Behavior. The mostserious shortcoming of the cross-section studies is that the explana-tory variable of interest is not independent of the taxpayer's behavior.The group of taxpayers under observation are all facing the same taxlaw. For a given filing status and family size, differences in marginaltax rates can result only from differences in the level of taxable in-come. This means that differences in tax rates among taxpayersreflect either differences in pretax income levels or differences in be-havior with respect to the form in which income is received, theamount of deductions claimed, or the realization of capital gains.

2. Lawrence B. Lindsey, Capital Gains: Rates, Realizations and Revenues, Working Paper No. 1893(National Bureau of Economic Research, Inc., April 1986).

APPENDIX B METHODOLOGICAL ISSUES 99

Thus, a negative relationship between realizations and marginal taxrates among taxpayers, often interpreted to mean that lower marginaltax rates cause more realizations, can have an alternative and equallyplausible interpretation. It could instead mean that taxpayers whoundertake behavior that lowers their tax rates also have more gains.

This problem of interpretation has two separate sources. The firstrelates to the direct causal relationship between capital gains realiza-tions and the marginal tax rate. As taxpayers realize more gains, theymay move into higher tax brackets. This tax bracket effect produces apositive statistical relationship between realizations and marginal taxrates that partially offsets the hypothesized negative causal linkrunning in the opposite direction from tax rates to realizations. With-out correction for this bias, statistical studies might underestimatethe degree to which realizations are negatively affected by high taxrates. The cross-section studies reviewed in Chapter II eliminated thissource of bias by using a "corrected" marginal tax rate measure that isindependent of the taxpayer's level of realizations.

The second source of bias is that other aspects of a taxpayer's be-havior—in particular, the taxpayer's choice among alternative invest-ment assets—can also affect the marginal tax rate. This source of biasprobably results in an overstatement of the effect of tax rates onrealized gains, but has not been corrected for in published work be-cause it is harder to identify directly. For example, consider a tax-payer who holds stocks with high dividend payout ratios and does notinvest in tax shelter activities. This taxpayer will have a higher mar-ginal tax rate, at the same level of economic income, than one whochooses a more risky strategy of holding growth stocks and tax shelterinvestments, and will also hold assets that generate much fewercapital gains. Thus, a comparison of the two taxpayers may produce anegative statistical relationship between tax rates and realized capi-tal gains that reflects differences in the portfolio holdings of the twotaxpayers, not a causal link between tax rates and the propensity torealize gains. More generally, since any difference in observed mar-ginal tax rates between two taxpayers with the same income level inthe same year reflects some differences in their behavior, the causallink between tax rates and gains in cross-section studies is highly am-biguous. One must assume, in effect, that the behavior that affectsmarginal tax rates has no independent influence on the level ofrealized capital gains in order to conclude that the estimated coeffi-


cient of the marginal tax rate term indicates how much realizationswill change in response to a legislated change in the tax rate.

Transitory versus Permanent Tax Rates. A second problem of inter-pretation is that some of the differences in marginal tax rates amongtaxpayers in any year reflect temporary instead of long-run differ-ences. A high-income taxpayer may have a temporarily low tax ratein any given year because of extraordinary medical expenses, tem-porary business losses, or a number of other circumstances. Tax-payers with varying levels of taxable income have an incentive torealize gains in those years when their tax rates are temporarily low.These timing effects do not, however, imply that realizations will bepermanently higher if the tax rate schedule is lowered. Thus, if someof the differences in tax rates among taxpayers are temporary, the es-timated permanent effect of lower marginal tax rates on realizationsin studies using data from one year is biased upward.

As noted in Chapter II, several studies have adjusted in part forthis problem by using panel data that follow the same taxpayers overseveral years. These studies use separate measures of "transitory"and "permanent" marginal tax rates, where the permanent tax rate isthe average of marginal tax rates in the year of observation and thetwo previous years.3/ The use of separate permanent and transitoryrates lowers the estimated tax rate effect, although there continues tobe an estimated "permanent" effect of lower marginal tax rates on re-alizations in these studies, as summarized in Chapter n.

Data Limited to Tax Returns. The advantage cross-section studiesenjoy of using individual taxpayer records is also a limitation becausetax return data omit some important explanatory information, amongthe most crucial of which are data on asset holdings. In theory, ac-crued gains should be included as a variable explaining differences inrealizations among taxpayers. While time-series studies can usewealth measures as approximations of accrued gains, the cross-sectionstudies cannot do this because no direct measure of wealth is reportedon tax returns.

3. See Gerald E. Auten and Charles T. Clotfelter, "Permanent Versus Transitory Tax Effects and theRealization of Capital Gains," Quarterly Journal of Economics, vol. 97 (November 1982), and Officeof the Secretary of the Treasury, Capital Gains Tax Reductions.


In place of wealth, cross-section studies must use "proxy" mea-sures of income that are reported on tax returns and are believed to becorrelated with wealth (and therefore with accrued capital gains).Dividends are used as a proxy for wealth in corporate equities, and in-come reported on tax returns from partnerships, proprietorships, andfarms is used as a proxy for the market value of other assets that giverise to capital gains. The tax return values are imperfect, however,because dividend-payout ratios differ among stocks held by differenttaxpayers and because reported net income from partnerships andproprietorships on tax returns is a very poor indicator of economic in-come.4/ The choice of an "incorrect" income measure can bias the esti-mate of the effect of marginal tax rates on realizations.

Other Issues. Other problems on which investigators have differedinclude the choice of which other variables to use in explaining capitalgains, the appropriate estimating technique, and the correct method ofweighting observations.5/ These methodological problems, and thedifferent ways they are handled, explain in part the wide range of taxrate responses estimated in cross-section studies.

ISSUES IN TIME-SERIES STUDIES

Time-series analyses can avoid the major problems of cross-sectionstudies, but at the expense of incurring different problems. The mainproblem of time-series analyses is the limited number of years of data;other problems are the aggregate nature of the data and difficulties indetermining the timing of tax changes.

4. As an alternative, the Treasury Department study used positive income only as a proxy for wealth.See Office of the Secretary of the Treasury, Capital Gains Tax Reductions.

5. The appropriate estimating technique is an issue because many taxpayers realize no capital gainsin any particular year. In such samples, the commonly used technique of ordinary least squaresregression gives biased estimates and normal tests of significance are inappropriate. One way tocorrect the problem is to use an alternative technique, such as Tobit analysis, that constrains thedependent variable. See Office of the Secretary of the Treasury, Capital Gains Tax Reductions.


Advantages Over Cross-Section Studies

One major problem of cross-section studies is that the tax rate is notindependent of the individual's behavior, and may actually be decidedsimultaneously with the individual's decision to realize gains. Thisproblem is largely avoided in time-series analysis because the majorchanges in marginal tax rates are caused by legislation and are out-side an individual's direct control .67

A second problem with cross-section analysis is that differencesamong individuals in a single year may be a poor guide to how any oneof them will respond to changes in his or her tax rates, particularlyhow slow or rapid that response will be. Since time-series studies arebased on the actual level of aggregate realizations, patterns of re-sponse spread out over more than one year may be detected.

A third advantage of aggregate time-series analysis over cross-section analysis is that data from many different sources can be used.For example, in this study, realizations and marginal tax rates fromtax returns are combined with corporate equity values reported by theFederal Reserve Board and with GNP, price level, and depreciationdata from the National Income and Product Accounts. Approximatingthese nontax variables from information on tax returns would sacri-fice accuracy.

Problems Unique to Time-Series Analysis

Few Years of Data. One major problem in time-series analysis,though not in cross-section studies, is the limited number of observa-tions. Tax return data on realizations and marginal tax rates used inthese studies go back only to 1954, and the most recent available tax

6. A form of endogeneity could be introduced in the construction of average tax rates, but this can beavoided by not using current year realizations in calculating marginal tax rates or in aggregatingrates among income classes. In this study, a standardized level of realizations is obtained for eachAGI class in each year, and that amount is used to calculate marginal tax rates and to aggregaterates among classes. A simpler alternative is to use the legislated maximum rate on capital gains.This alternative eliminates endogeneity, but at the cost of suppressing relevant data on changes intax rates faced by taxpayers below the highest-income group.


return data are for 1985.77 That gives 32 observations, effectivelylimiting the analysis to 5 to 10 independent variables.

The available data are actually even more limited than the num-ber of observations would suggest. To begin with, major changes intax rates have occurred only five times in the 32-year period, and onlythree distinct periods stand out. As Figure 4 in Chapter HI shows,rates were steady through 1969, rose through 1978, and were lowagain from 1979 on. Moreover, many features of the economy changeonly slowly from one year to the next causing many of the explanatoryvariables to be highly correlated over the 32 years or for importantsegments of those years. For example, the era of high tax rates ongains coincides with an era of high inflation and low appreciation ofstock prices. Similarly, the era of rapidly rising realizations since1978 coincides with declining tax rates, a rising stock market in mostyears, the growth of accelerated depreciation and tax shelters, thegrowth of corporate takeovers and leveraged buy-outs, a decline inbrokerage commissions, and the imposition of reporting requirementson stock sales. It is not possible to distinguish completely the separateinfluence of tax rate changes when the changes are few and severalother changes happen at about the same time.

Aggregated Data. Most estimated time-series equations are specifiedas if the data reflect a single representative taxpayer over the entiresample period. The composition of the taxpaying population, however,has changed substantially over the past 30 years. Since 1954, thepopulation has become older, households have come to hold a greatershare of their equity indirectly in pension funds and mutual funds,and the ownership of corporate equity, as estimated by dividend in-come, appears to have become less concentrated among the highest-income groups. Each of these changes could cause shifts in the re-sponse of realizations to tax rates, as could many other changes in thepopulation, but their effects are not captured in the aggregate time-series models.

Even if the population were completely static, as assumed by theaggregate models, the aggregate data do not precisely reflect the sumof the relevant data for individual taxpayers. Realizations are often

7. Data for years before the mid-1970s are available only in printed form, precluding the constructionof variables in any way not already tabulated.


used logarithmically or as a ratio to other aggregate variables. Evenwhen the log form is appropriate, the aggregation introduces some im-precision because the logarithm of total realizations is not equal to thesum of the logarithms of realizations for each individual taxpayer.The choice of a representative marginal tax rate illustrates a differenttype of aggregation problem. Ideally, one would want to use a gains-weighted average of marginal tax rates across all tax returns. Thiscannot be closely approximated in the absence of data on a largesample of individual returns. In some studies, the maximum capitalgains tax rate is used to represent tax rates for all taxpayers.8/ In thisstudy, tax rates are imputed to classes of taxpayers at different AGIlevels and then a weighted average marginal tax rate is calculated.This approach represents a wider range of taxpayers than the use ofthe maximum rate alone, and can capture the effects of changes inrates below the top rate. Because individual records are unavailable,however, the calculation of the average marginal rate for each AGIclass is necessarily imprecise. No account can be taken of filingstatus; those with realized gains cannot be isolated from other returns;and the within-class distribution of gains and marginal tax rates isunavailable.

Timing of Tax Changes. A final time-series problem is the difficulty ofrepresenting the time at which legislated tax changes begin to affectrealizations. When legislative changes take effect in the middle of ayear, as in 1978 and 1981, it is not possible from available tax data todetermine what portion of a year's gains occurred after the change.Furthermore, tax legislation often takes a long time to be enacted intolaw, so that the prospect of a change in law may affect realizationsbefore the change is actually enacted. For example, the tax changesenacted in 1981 were discussed in the 1980 election campaign andeven before, so that it is impossible to know just when the changes, ortheir prospect, may have altered behavior. The opportunity to observethe timing of responses to tax changes, even if imprecisely, is onepotential advantage of the use of time-series analysis that may bebetter utilized when post-tax-reform data become available. (Anequation that attempts to estimate the impact of anticipated taxchanges is presented in Appendix A.)

8. Office of the Secretary of the Treasury, Capital Gains Tax Reductions.


COMBINED TIME-SERIES AND CROSS-SECTION ANALYSIS

An ideal but currently unobtainable data base for studying the effectof marginal tax rates on the realizations of capital gains would be onehaving many observations as used in cross-section studies, but thatalso allowed the researcher to observe the effects of changes in the taxlaw over time. Such a data file might consist of a large sample ofrecords of individual taxpayers that included data on capital assetstransactions and followed the same taxpayers over an extended timeperiod. Some studies have made use of such "panel" data, but the datahave generally not encompassed years when there have been majorchanges in the tax law.

The major use to date of taxpayer panel data in studying capitalgains has been to modify slightly the results of single-year cross-section studies. As described above, the data from adjoining yearshave been used to develop separate measures of "transitory" and"permanent" marginal tax rates for individual taxpayers, with thepermanent rates constructed as an average of marginal tax rates overseveral years. In all these studies, the dependent variable is stillrealizations in a single year; no effort has been made to estimate theactual response of individual taxpayers to changes in the tax law overtime.

Work with panel data would appear to be a fruitful course ofstudy. At the present time, however, no taxpayer panel that includesthe desired information on sales of capital assets and covers recentyears in which capital gains taxation has been changing is publiclyavailable.

As noted in the literature review in Chapter II, Lindsey hasperformed a pooled cross-section time-series study, which takes asunits of observation six separate taxpayer groups over a period of 18years, and found a significant and large negative relationship betweenrealized long-term capital gains and marginal tax rates on gains. Thetaxpayer groups are defined by ranges of AGI that are fixed in ab-solute dollars; thus, they do not include the same taxpayers in dif-ferent years and also each group represents a different percentage ofthe total taxpaying population in different years. As discussed inAppendix A, aggregate time-series equations using variables similarto those used by Lindsey do not reveal a negative statistical relation-


ship between capital gains and marginal tax rates. It would appearthat Lindsey's results reflect primarily the effects of differences in taxrates and realizations among groups of taxpayers instead of the effectsof changes in tax rates over time.

The use of taxpayer groups as the unit of observation raises diffi-cult problems of interpretation in addition to those in cross-sectionstudies using individual data. When using income groups, the findingof a negative relationship between marginal tax rates and gainsrealizations occurs because those groups with the lowest marginal taxrates—that is, the lowest-income groups—realize more gains thanwould be expected given the amounts of dividend income, interest in-come, and income from other capital assets received by taxpayers inthose groups. As with other studies, one must ask whether the higherrealized gains result because of the independent effect of lowermarginal tax rates on gains or because of some other differencebetween the observed taxpayers. In the case of cross-section studies ofindividuals, one can control for the effects of many other taxpayervariables, including the taxpayer's total income. When using broadincome groups as units of observation, however, one cannot examinethe effects of marginal tax rates on the behavior of taxpayers withsimilar incomes. Thus, while the findings can be suggestive of apotentially large inverse relationship between tax rates and gains,they could also simply be capturing some other relationship amongsources of capital income received by different income groups.

how capital gains tax rates affect revenues: the historical

Documents